From da663622f531ffd0c66e8f309d4faf301eb8b503 Mon Sep 17 00:00:00 2001
From: Julien Breton <julien.breton@irit.fr>
Date: Tue, 9 Jul 2024 15:25:21 +0200
Subject: [PATCH] perform legal bert
---
.gitignore | 4 +
models/download_model.py | 3 +-
.../finetuned-legal-camembert.py | 0
.../finetuning-legal-bert-classifier.ipynb | 2222 +++++++++++++++++
modules/camembert/camembert-classifier.ipynb | 662 -----
results/BERT/legal_bert.png | Bin 0 -> 32101 bytes
results/{Hybridation => BERT}/performance.png | Bin
temp.ipynb | 125 -
8 files changed, 2228 insertions(+), 788 deletions(-)
rename modules/{camembert => bert}/finetuned-legal-camembert.py (100%)
create mode 100644 modules/bert/finetuning-legal-bert-classifier.ipynb
delete mode 100644 modules/camembert/camembert-classifier.ipynb
create mode 100644 results/BERT/legal_bert.png
rename results/{Hybridation => BERT}/performance.png (100%)
delete mode 100644 temp.ipynb
diff --git a/.gitignore b/.gitignore
index 8142d36..c722019 100644
--- a/.gitignore
+++ b/.gitignore
@@ -176,6 +176,10 @@ models/CamemBERT-base/*
!models/CamemBERT-base/.gitkeep
models/Fine-tuned_CamemBERT-base/*
!models/Fine-tuned_CamemBERT-base/.gitkeep
+models/LegalCamemBERT-base/*
+!models/LegalCamemBERT-base/.gitkeep
+models/Fine-tuned_LegalCamemBERT-base/*
+!models/Fine-tuned_LegalCamemBERT-base/.gitkeep
modules/llm/Mixtral-8x7b/results/*
modules/llm/Mixtral-8x7b/.lock_preprocessing
diff --git a/models/download_model.py b/models/download_model.py
index 26ac063..72530a1 100644
--- a/models/download_model.py
+++ b/models/download_model.py
@@ -3,4 +3,5 @@ from huggingface_hub import snapshot_download
#snapshot_download(repo_id="mistralai/Mixtral-8x7B-Instruct-v0.1", cache_dir="../temp", local_dir="./Mixtral-8x7B-Instruct-v0.1")
#snapshot_download(repo_id="mistralai/Mistral-7B-Instruct-v0.2", cache_dir="../temp", local_dir="./Mistral-7B-Instruct-v0.2")
#snapshot_download(repo_id="152334H/miqu-1-70b-sf", cache_dir="../temp", local_dir="./Miqu-1-70b")
-snapshot_download(repo_id="camembert/camembert-base", cache_dir="../temp", local_dir="./CamemBERT-base")
\ No newline at end of file
+#snapshot_download(repo_id="camembert/camembert-base", cache_dir="../temp", local_dir="./CamemBERT-base")
+snapshot_download(repo_id="maastrichtlawtech/legal-camembert-base", cache_dir="../temp", local_dir="./LegalCamemBERT-base")
\ No newline at end of file
diff --git a/modules/camembert/finetuned-legal-camembert.py b/modules/bert/finetuned-legal-camembert.py
similarity index 100%
rename from modules/camembert/finetuned-legal-camembert.py
rename to modules/bert/finetuned-legal-camembert.py
diff --git a/modules/bert/finetuning-legal-bert-classifier.ipynb b/modules/bert/finetuning-legal-bert-classifier.ipynb
new file mode 100644
index 0000000..c104075
--- /dev/null
+++ b/modules/bert/finetuning-legal-bert-classifier.ipynb
@@ -0,0 +1,2222 @@
+{
+ "cells": [
+ {
+ "cell_type": "markdown",
+ "id": "8ced2e3ca31fb46c",
+ "metadata": {},
+ "source": [
+ "# Dataset"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "id": "757a8bf026156e77",
+ "metadata": {
+ "ExecuteTime": {
+ "end_time": "2024-06-27T14:45:25.410725Z",
+ "start_time": "2024-06-27T14:45:25.404357Z"
+ }
+ },
+ "outputs": [],
+ "source": [
+ "tag2id = {'action': 1, 'actor': 2, 'artifact': 3, 'condition': 4, 'location': 5, 'modality': 6, 'reference': 7, 'time': 8}\n",
+ "id2tag = {v:k for k, v in tag2id.items()}"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 2,
+ "id": "be3a4c320f9d4a5",
+ "metadata": {
+ "ExecuteTime": {
+ "end_time": "2024-06-27T14:45:35.889919Z",
+ "start_time": "2024-06-27T14:45:35.885841Z"
+ },
+ "editable": true,
+ "slideshow": {
+ "slide_type": ""
+ },
+ "tags": []
+ },
+ "outputs": [],
+ "source": [
+ "label2id = {\n",
+ " 'O': 0,\n",
+ " **{f'{k}': v for k, v in tag2id.items()}\n",
+ "}\n",
+ "\n",
+ "id2label = {v:k for k, v in label2id.items()}"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 3,
+ "id": "2aa2fefac95e7f04",
+ "metadata": {
+ "ExecuteTime": {
+ "end_time": "2024-06-27T14:45:46.507242Z",
+ "start_time": "2024-06-27T14:45:38.632569Z"
+ }
+ },
+ "outputs": [],
+ "source": [
+ "from datasets import Dataset\n",
+ "train_ds = Dataset.from_json(\"../../data/annotations.train.jsonlines\")\n",
+ "val_ds = Dataset.from_json(\"../../data/annotations.eval.jsonlines\")"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 4,
+ "id": "9e0a21356e7701a1",
+ "metadata": {
+ "ExecuteTime": {
+ "end_time": "2024-06-27T14:45:52.188551Z",
+ "start_time": "2024-06-27T14:45:52.185648Z"
+ }
+ },
+ "outputs": [],
+ "source": [
+ "modelId = '../../models/LegalCamemBERT-base'"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "66e00d5a79a66753",
+ "metadata": {},
+ "source": [
+ "# Tokenization"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 5,
+ "id": "e6459259f5ab2d98",
+ "metadata": {
+ "ExecuteTime": {
+ "end_time": "2024-06-27T14:46:02.189231Z",
+ "start_time": "2024-06-27T14:45:56.737350Z"
+ }
+ },
+ "outputs": [],
+ "source": [
+ "from transformers import AutoTokenizer\n",
+ "tokenizer = AutoTokenizer.from_pretrained(modelId)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 6,
+ "id": "8c96680645f077fb",
+ "metadata": {
+ "ExecuteTime": {
+ "end_time": "2024-06-27T14:46:08.307140Z",
+ "start_time": "2024-06-27T14:46:08.301542Z"
+ }
+ },
+ "outputs": [],
+ "source": [
+ "def get_token_role_in_span(token_start: int, token_end: int, span_start: int, span_end: int):\n",
+ " \"\"\"\n",
+ " Check if the token is inside a span.\n",
+ " Args:\n",
+ " - token_start, token_end: Start and end offset of the token\n",
+ " - span_start, span_end: Start and end of the span\n",
+ " Returns:\n",
+ " - \"B\" if beginning\n",
+ " - \"I\" if inner\n",
+ " - \"O\" if outer\n",
+ " - \"N\" if not valid token (like <SEP>, <CLS>, <UNK>)\n",
+ " \"\"\"\n",
+ " if token_end <= token_start:\n",
+ " return \"N\"\n",
+ " if token_start < span_start or token_end > span_end:\n",
+ " return \"O\"\n",
+ " else:\n",
+ " return \"I\"\n",
+ "\n",
+ "MAX_LENGTH = 256\n",
+ "\n",
+ "def tokenize_and_adjust_labels(sample):\n",
+ " \"\"\"\n",
+ " Args:\n",
+ " - sample (dict): {\"id\": \"...\", \"text\": \"...\", \"tags\": [{\"start\": ..., \"end\": ..., \"tag\": ...}, ...]\n",
+ " Returns:\n",
+ " - The tokenized version of `sample` and the labels of each token.\n",
+ " \"\"\"\n",
+ " # Tokenize the text, keep the start and end positions of tokens with `return_offsets_mapping` option\n",
+ " # Use max_length and truncation to ajust the text length\n",
+ " tokenized = tokenizer(sample[\"text\"],\n",
+ " return_offsets_mapping=True,\n",
+ " padding=\"max_length\",\n",
+ " max_length=MAX_LENGTH,\n",
+ " truncation=True)\n",
+ "\n",
+ " # We are doing a multilabel classification task at each token, we create a list of size len(label2id)=13 \n",
+ " # for the 13 labels\n",
+ " labels = [[0 for _ in label2id.keys()] for _ in range(MAX_LENGTH)]\n",
+ "\n",
+ " # Scan all the tokens and spans, assign 1 to the corresponding label if the token lies at the beginning\n",
+ " # or inside the spans\n",
+ " for (token_start, token_end), token_labels in zip(tokenized[\"offset_mapping\"], labels):\n",
+ " #print(token_start, token_end)\n",
+ " for span in sample[\"tags\"]:\n",
+ " role = get_token_role_in_span(token_start, token_end, span[\"start\"], span[\"end\"])\n",
+ " if role == \"I\":\n",
+ " token_labels[label2id[f\"{span['tag']}\"]] = 1\n",
+ "\n",
+ " return {**tokenized, \"labels\": labels}"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 7,
+ "id": "53310845f13e9d70",
+ "metadata": {
+ "ExecuteTime": {
+ "end_time": "2024-06-27T14:46:13.910535Z",
+ "start_time": "2024-06-27T14:46:13.689422Z"
+ }
+ },
+ "outputs": [],
+ "source": [
+ "tokenized_train_ds = train_ds.map(tokenize_and_adjust_labels, remove_columns=train_ds.column_names)\n",
+ "tokenized_val_ds = val_ds.map(tokenize_and_adjust_labels, remove_columns=val_ds.column_names)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 8,
+ "id": "a654180a-a536-43fa-8984-c7b87b419f93",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "--------Token---------|--------Offset----------|--------Labels----------\n",
+ " <s> | [0, 0] | []\n",
+ " Est | [0, 3] | ['action']\n",
+ " puni | [4, 8] | ['action']\n",
+ " des | [9, 12] | ['action']\n",
+ " mêmes | [13, 18] | ['action']\n",
+ " peines | [19, 25] | ['action']\n",
+ " le | [26, 28] | []\n",
+ " fait | [29, 33] | []\n",
+ " de | [34, 36] | []\n",
+ " tolérer | [37, 44] | []\n",
+ " comme | [45, 50] | []\n",
+ " propriétaire | [51, 63] | ['actor']\n",
+ " ou | [64, 66] | ['actor']\n",
+ " détenteur | [67, 76] | ['actor']\n",
+ " la | [77, 79] | []\n",
+ " mise | [80, 84] | []\n",
+ " en | [85, 87] | []\n",
+ " circulation | [88, 99] | []\n",
+ " d | [100, 101] | []\n",
+ " | [102, 103] | []\n",
+ " ' | [102, 103] | []\n",
+ " un | [104, 106] | []\n",
+ " véhicule | [107, 115] | []\n",
+ " sur | [116, 119] | ['location']\n",
+ " les | [120, 123] | ['location']\n",
+ " voies | [124, 129] | ['location']\n",
+ " publiques | [130, 139] | ['location']\n",
+ " par | [140, 143] | []\n",
+ " une | [144, 147] | ['actor']\n",
+ " personne | [148, 156] | ['actor']\n",
+ " non | [157, 160] | ['condition']\n",
+ " titulaire | [161, 170] | ['condition']\n",
+ " d | [171, 172] | ['condition']\n",
+ " | [173, 174] | ['condition']\n",
+ " ' | [173, 174] | ['condition']\n",
+ " un | [175, 177] | ['condition']\n",
+ " permis | [178, 184] | ['condition']\n",
+ " de | [185, 187] | ['condition']\n",
+ " conduire | [188, 196] | ['condition']\n",
+ " valable | [197, 204] | ['condition']\n",
+ " | [205, 206] | []\n",
+ " . | [205, 206] | []\n",
+ " </s> | [0, 0] | []\n"
+ ]
+ }
+ ],
+ "source": [
+ "sample = tokenized_train_ds[0]\n",
+ "print(\"--------Token---------|--------Offset----------|--------Labels----------\")\n",
+ "for token_id, token_labels, offset in zip(sample[\"input_ids\"], sample[\"labels\"], sample[\"offset_mapping\"]):\n",
+ " # Decode the token_id into text\n",
+ " token_text = tokenizer.decode(token_id)\n",
+ " \n",
+ " # Retrieve all the indices corresponding to the \"1\" at each token, decode them to label name\n",
+ " labels = [id2label[label_index] for label_index, value in enumerate(token_labels) if value==1]\n",
+ " \n",
+ " # Decode those indices into label name\n",
+ " print(f\" {token_text:20} | {offset} | {labels}\")\n",
+ " \n",
+ " # Finish when we meet the end of sentence.\n",
+ " if token_text == \"</s>\": \n",
+ " break"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 9,
+ "id": "6990d89800dbb440",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "from transformers import DataCollatorWithPadding\n",
+ "data_collator = DataCollatorWithPadding(tokenizer, padding=True)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "668dcf9750404d1c",
+ "metadata": {},
+ "source": [
+ "# Adapt the model"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 10,
+ "id": "7bd0cddab7ddb448",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "import numpy as np\n",
+ "from sklearn.metrics import multilabel_confusion_matrix\n",
+ "\n",
+ "n_labels = len(id2label)\n",
+ "\n",
+ "def divide(a: int, b: int):\n",
+ " return a / b if b > 0 else 0\n",
+ "\n",
+ "def compute_metrics(p):\n",
+ " \"\"\"\n",
+ " Customize the `compute_metrics` of `transformers`\n",
+ " Args:\n",
+ " - p (tuple): 2 numpy arrays: predictions and true_labels\n",
+ " Returns:\n",
+ " - metrics (dict): f1 score on \n",
+ " \"\"\"\n",
+ " # (1)\n",
+ " predictions, true_labels = p\n",
+ "\n",
+ " # (2)\n",
+ " predicted_labels = np.where(predictions > 0, np.ones(predictions.shape), np.zeros(predictions.shape))\n",
+ " metrics = {}\n",
+ "\n",
+ " # (3)\n",
+ " cm = multilabel_confusion_matrix(true_labels.reshape(-1, n_labels), predicted_labels.reshape(-1, n_labels))\n",
+ "\n",
+ " # (4) \n",
+ " for label_idx, matrix in enumerate(cm):\n",
+ " if label_idx == 0:\n",
+ " continue # We don't care about the label \"O\"\n",
+ " tp, fp, fn = matrix[1, 1], matrix[0, 1], matrix[1, 0]\n",
+ " precision = divide(tp, tp + fp)\n",
+ " recall = divide(tp, tp + fn)\n",
+ " f1 = divide(2 * precision * recall, precision + recall)\n",
+ " metrics[f\"f1_{id2label[label_idx]}\"] = f1\n",
+ "\n",
+ " # (5)\n",
+ " macro_f1 = sum(list(metrics.values())) / (n_labels - 1)\n",
+ " metrics[\"macro_f1\"] = macro_f1\n",
+ "\n",
+ " return metrics"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 11,
+ "id": "ea5d16f59728e2b9",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "from transformers import AutoModelForTokenClassification, TrainingArguments, Trainer, TrainerCallback\n",
+ "from transformers import RobertaPreTrainedModel, RobertaModel\n",
+ "from transformers.utils import (\n",
+ " add_code_sample_docstrings,\n",
+ " add_start_docstrings,\n",
+ " add_start_docstrings_to_model_forward,\n",
+ " logging,\n",
+ " replace_return_docstrings,\n",
+ ")\n",
+ "from transformers.models.roberta.modeling_roberta import (\n",
+ " ROBERTA_INPUTS_DOCSTRING,\n",
+ " ROBERTA_START_DOCSTRING,\n",
+ " RobertaEmbeddings,\n",
+ ")\n",
+ "from typing import Optional, Union, Tuple\n",
+ "from transformers.modeling_outputs import TokenClassifierOutput\n",
+ "import torch\n",
+ "from torch import nn\n",
+ "import matplotlib.pyplot as plt\n",
+ "from transformers.trainer_utils import IntervalStrategy\n",
+ "import pandas as pd\n",
+ "\n",
+ "class RobertaForSpanCategorization(RobertaPreTrainedModel):\n",
+ " _keys_to_ignore_on_load_unexpected = [r\"pooler\"]\n",
+ " _keys_to_ignore_on_load_missing = [r\"position_ids\"]\n",
+ "\n",
+ " def __init__(self, config):\n",
+ " super().__init__(config)\n",
+ " self.num_labels = config.num_labels\n",
+ " self.roberta = RobertaModel(config, add_pooling_layer=False)\n",
+ " classifier_dropout = (\n",
+ " config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob\n",
+ " )\n",
+ " self.dropout = nn.Dropout(classifier_dropout)\n",
+ " self.classifier = nn.Linear(config.hidden_size, config.num_labels)\n",
+ " # Initialize weights and apply final processing\n",
+ " self.post_init()\n",
+ "\n",
+ " @add_start_docstrings_to_model_forward(ROBERTA_INPUTS_DOCSTRING.format(\"batch_size, sequence_length\"))\n",
+ " def forward(\n",
+ " self,\n",
+ " input_ids: Optional[torch.LongTensor] = None,\n",
+ " attention_mask: Optional[torch.FloatTensor] = None,\n",
+ " token_type_ids: Optional[torch.LongTensor] = None,\n",
+ " position_ids: Optional[torch.LongTensor] = None,\n",
+ " head_mask: Optional[torch.FloatTensor] = None,\n",
+ " inputs_embeds: Optional[torch.FloatTensor] = None,\n",
+ " labels: Optional[torch.LongTensor] = None,\n",
+ " output_attentions: Optional[bool] = None,\n",
+ " output_hidden_states: Optional[bool] = None,\n",
+ " return_dict: Optional[bool] = None,\n",
+ " ) -> Union[Tuple[torch.Tensor], TokenClassifierOutput]:\n",
+ " r\"\"\"\n",
+ " labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):\n",
+ " Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`.\n",
+ " \"\"\"\n",
+ " return_dict = return_dict if return_dict is not None else self.config.use_return_dict\n",
+ " outputs = self.roberta(\n",
+ " input_ids,\n",
+ " attention_mask=attention_mask,\n",
+ " token_type_ids=token_type_ids,\n",
+ " position_ids=position_ids,\n",
+ " head_mask=head_mask,\n",
+ " inputs_embeds=inputs_embeds,\n",
+ " output_attentions=output_attentions,\n",
+ " output_hidden_states=output_hidden_states,\n",
+ " return_dict=return_dict,\n",
+ " )\n",
+ " sequence_output = outputs[0]\n",
+ " sequence_output = self.dropout(sequence_output)\n",
+ " logits = self.classifier(sequence_output)\n",
+ "\n",
+ " loss = None\n",
+ " if labels is not None:\n",
+ " loss_fct = nn.BCEWithLogitsLoss()\n",
+ " loss = loss_fct(logits, labels.float())\n",
+ " if not return_dict:\n",
+ " output = (logits,) + outputs[2:]\n",
+ " return ((loss,) + output) if loss is not None else output\n",
+ " return TokenClassifierOutput(\n",
+ " loss=loss,\n",
+ " logits=logits,\n",
+ " hidden_states=outputs.hidden_states,\n",
+ " attentions=outputs.attentions,\n",
+ " )\n",
+ "\n",
+ "class TrainingMetricsCallback(TrainerCallback):\n",
+ " def __init__(self):\n",
+ " self.macro_f1 = []\n",
+ " self.steps = []\n",
+ " self.counter = 0\n",
+ "\n",
+ " def on_evaluate(self, args, state, control, metrics=None, **kwargs):\n",
+ " if metrics is not None:\n",
+ " if 'eval_macro_f1' in metrics:\n",
+ " self.macro_f1.append(metrics['eval_macro_f1'])\n",
+ " self.counter += 1\n",
+ " self.steps.append(self.counter)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "77f4fc68394aa754",
+ "metadata": {},
+ "source": [
+ "# Fine-tuning"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 12,
+ "id": "79161ed938cad895",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "training_args = TrainingArguments(\n",
+ " output_dir=\"./models/fine_tune_bert_output_span_cat\",\n",
+ " evaluation_strategy=\"epoch\",\n",
+ " learning_rate=2.5e-4,\n",
+ " per_device_train_batch_size=16,\n",
+ " per_device_eval_batch_size=16,\n",
+ " num_train_epochs=100,\n",
+ " weight_decay=0.01,\n",
+ " logging_steps = 100,\n",
+ " save_strategy='epoch',\n",
+ " save_total_limit=2,\n",
+ " load_best_model_at_end=True,\n",
+ " metric_for_best_model='macro_f1',\n",
+ " log_level='critical',\n",
+ " seed=12345\n",
+ ")"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 13,
+ "id": "931792b554582a9f",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ " <div>\n",
+ " \n",
+ " <progress value='1300' max='1300' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
+ " [1300/1300 09:06, Epoch 100/100]\n",
+ " </div>\n",
+ " <table border=\"1\" class=\"dataframe\">\n",
+ " <thead>\n",
+ " <tr style=\"text-align: left;\">\n",
+ " <th>Epoch</th>\n",
+ " <th>Training Loss</th>\n",
+ " <th>Validation Loss</th>\n",
+ " <th>F1 Action</th>\n",
+ " <th>F1 Actor</th>\n",
+ " <th>F1 Artifact</th>\n",
+ " <th>F1 Condition</th>\n",
+ " <th>F1 Location</th>\n",
+ " <th>F1 Modality</th>\n",
+ " <th>F1 Reference</th>\n",
+ " <th>F1 Time</th>\n",
+ " <th>Macro F1</th>\n",
+ " </tr>\n",
+ " </thead>\n",
+ " <tbody>\n",
+ " <tr>\n",
+ " <td>1</td>\n",
+ " <td>No log</td>\n",
+ " <td>0.356145</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.000000</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>2</td>\n",
+ " <td>No log</td>\n",
+ " <td>0.279355</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.000000</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>3</td>\n",
+ " <td>No log</td>\n",
+ " <td>0.221275</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.000000</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>4</td>\n",
+ " <td>No log</td>\n",
+ " <td>0.178497</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.000000</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>5</td>\n",
+ " <td>No log</td>\n",
+ " <td>0.149304</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.000000</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>6</td>\n",
+ " <td>No log</td>\n",
+ " <td>0.127901</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.000000</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>7</td>\n",
+ " <td>No log</td>\n",
+ " <td>0.114968</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.000000</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>8</td>\n",
+ " <td>0.213000</td>\n",
+ " <td>0.102522</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.000000</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>9</td>\n",
+ " <td>0.213000</td>\n",
+ " <td>0.097100</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.176965</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.022121</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>10</td>\n",
+ " <td>0.213000</td>\n",
+ " <td>0.090857</td>\n",
+ " <td>0.346816</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.779005</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.140728</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>11</td>\n",
+ " <td>0.213000</td>\n",
+ " <td>0.086320</td>\n",
+ " <td>0.157806</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.771041</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.116106</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>12</td>\n",
+ " <td>0.213000</td>\n",
+ " <td>0.083295</td>\n",
+ " <td>0.398428</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.749960</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.143549</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>13</td>\n",
+ " <td>0.213000</td>\n",
+ " <td>0.082080</td>\n",
+ " <td>0.347058</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.779387</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.140806</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>14</td>\n",
+ " <td>0.213000</td>\n",
+ " <td>0.079545</td>\n",
+ " <td>0.432345</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.787969</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.152539</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>15</td>\n",
+ " <td>0.213000</td>\n",
+ " <td>0.076363</td>\n",
+ " <td>0.371108</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.781354</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.144058</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>16</td>\n",
+ " <td>0.066000</td>\n",
+ " <td>0.075020</td>\n",
+ " <td>0.554452</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.759754</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.164276</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>17</td>\n",
+ " <td>0.066000</td>\n",
+ " <td>0.076133</td>\n",
+ " <td>0.518700</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.783399</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.162762</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>18</td>\n",
+ " <td>0.066000</td>\n",
+ " <td>0.072838</td>\n",
+ " <td>0.531309</td>\n",
+ " <td>0.197055</td>\n",
+ " <td>0</td>\n",
+ " <td>0.786690</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.540024</td>\n",
+ " <td>0</td>\n",
+ " <td>0.256885</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>19</td>\n",
+ " <td>0.066000</td>\n",
+ " <td>0.075695</td>\n",
+ " <td>0.445422</td>\n",
+ " <td>0.035851</td>\n",
+ " <td>0</td>\n",
+ " <td>0.733345</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.315789</td>\n",
+ " <td>0</td>\n",
+ " <td>0.191301</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>20</td>\n",
+ " <td>0.066000</td>\n",
+ " <td>0.072801</td>\n",
+ " <td>0.506315</td>\n",
+ " <td>0.060453</td>\n",
+ " <td>0</td>\n",
+ " <td>0.762457</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.600897</td>\n",
+ " <td>0</td>\n",
+ " <td>0.241265</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>21</td>\n",
+ " <td>0.066000</td>\n",
+ " <td>0.072491</td>\n",
+ " <td>0.466798</td>\n",
+ " <td>0.264026</td>\n",
+ " <td>0</td>\n",
+ " <td>0.779796</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.583524</td>\n",
+ " <td>0</td>\n",
+ " <td>0.261768</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>22</td>\n",
+ " <td>0.066000</td>\n",
+ " <td>0.068607</td>\n",
+ " <td>0.531232</td>\n",
+ " <td>0.440191</td>\n",
+ " <td>0.034570</td>\n",
+ " <td>0.786928</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.707031</td>\n",
+ " <td>0</td>\n",
+ " <td>0.312494</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>23</td>\n",
+ " <td>0.066000</td>\n",
+ " <td>0.067250</td>\n",
+ " <td>0.503735</td>\n",
+ " <td>0.086420</td>\n",
+ " <td>0.001042</td>\n",
+ " <td>0.797254</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.713208</td>\n",
+ " <td>0.208232</td>\n",
+ " <td>0.288736</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>24</td>\n",
+ " <td>0.039600</td>\n",
+ " <td>0.069559</td>\n",
+ " <td>0.518097</td>\n",
+ " <td>0.466793</td>\n",
+ " <td>0.047141</td>\n",
+ " <td>0.787783</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.608506</td>\n",
+ " <td>0.204489</td>\n",
+ " <td>0.329101</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>25</td>\n",
+ " <td>0.039600</td>\n",
+ " <td>0.068075</td>\n",
+ " <td>0.486739</td>\n",
+ " <td>0.497164</td>\n",
+ " <td>0.061265</td>\n",
+ " <td>0.774566</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.736209</td>\n",
+ " <td>0.555336</td>\n",
+ " <td>0.388910</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>26</td>\n",
+ " <td>0.039600</td>\n",
+ " <td>0.065591</td>\n",
+ " <td>0.526296</td>\n",
+ " <td>0.539711</td>\n",
+ " <td>0.176184</td>\n",
+ " <td>0.803005</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.743191</td>\n",
+ " <td>0.538232</td>\n",
+ " <td>0.415827</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>27</td>\n",
+ " <td>0.039600</td>\n",
+ " <td>0.073091</td>\n",
+ " <td>0.450490</td>\n",
+ " <td>0.438735</td>\n",
+ " <td>0.021572</td>\n",
+ " <td>0.770273</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.600897</td>\n",
+ " <td>0.234597</td>\n",
+ " <td>0.314570</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>28</td>\n",
+ " <td>0.039600</td>\n",
+ " <td>0.067930</td>\n",
+ " <td>0.492051</td>\n",
+ " <td>0.509091</td>\n",
+ " <td>0.106906</td>\n",
+ " <td>0.803842</td>\n",
+ " <td>0</td>\n",
+ " <td>0</td>\n",
+ " <td>0.674923</td>\n",
+ " <td>0.699515</td>\n",
+ " <td>0.410791</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>29</td>\n",
+ " <td>0.039600</td>\n",
+ " <td>0.064891</td>\n",
+ " <td>0.517025</td>\n",
+ " <td>0.592593</td>\n",
+ " <td>0.030334</td>\n",
+ " <td>0.820184</td>\n",
+ " <td>0</td>\n",
+ " <td>0.013333</td>\n",
+ " <td>0.713996</td>\n",
+ " <td>0.685200</td>\n",
+ " <td>0.421583</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>30</td>\n",
+ " <td>0.039600</td>\n",
+ " <td>0.063571</td>\n",
+ " <td>0.558310</td>\n",
+ " <td>0.576125</td>\n",
+ " <td>0.140530</td>\n",
+ " <td>0.803007</td>\n",
+ " <td>0</td>\n",
+ " <td>0.215569</td>\n",
+ " <td>0.723773</td>\n",
+ " <td>0.666118</td>\n",
+ " <td>0.460429</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>31</td>\n",
+ " <td>0.025000</td>\n",
+ " <td>0.064738</td>\n",
+ " <td>0.525939</td>\n",
+ " <td>0.589438</td>\n",
+ " <td>0.172210</td>\n",
+ " <td>0.818668</td>\n",
+ " <td>0</td>\n",
+ " <td>0.482412</td>\n",
+ " <td>0.718967</td>\n",
+ " <td>0.702580</td>\n",
+ " <td>0.501277</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>32</td>\n",
+ " <td>0.025000</td>\n",
+ " <td>0.065563</td>\n",
+ " <td>0.549824</td>\n",
+ " <td>0.636741</td>\n",
+ " <td>0.091671</td>\n",
+ " <td>0.806373</td>\n",
+ " <td>0</td>\n",
+ " <td>0.548077</td>\n",
+ " <td>0.687243</td>\n",
+ " <td>0.676329</td>\n",
+ " <td>0.499532</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>33</td>\n",
+ " <td>0.025000</td>\n",
+ " <td>0.062232</td>\n",
+ " <td>0.558872</td>\n",
+ " <td>0.690293</td>\n",
+ " <td>0.130246</td>\n",
+ " <td>0.823770</td>\n",
+ " <td>0.153846</td>\n",
+ " <td>0.561905</td>\n",
+ " <td>0.744412</td>\n",
+ " <td>0.719551</td>\n",
+ " <td>0.547862</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>34</td>\n",
+ " <td>0.025000</td>\n",
+ " <td>0.063367</td>\n",
+ " <td>0.542108</td>\n",
+ " <td>0.695394</td>\n",
+ " <td>0.170620</td>\n",
+ " <td>0.828333</td>\n",
+ " <td>0.253968</td>\n",
+ " <td>0.583333</td>\n",
+ " <td>0.734452</td>\n",
+ " <td>0.733387</td>\n",
+ " <td>0.567700</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>35</td>\n",
+ " <td>0.025000</td>\n",
+ " <td>0.061890</td>\n",
+ " <td>0.562546</td>\n",
+ " <td>0.707951</td>\n",
+ " <td>0.167960</td>\n",
+ " <td>0.821657</td>\n",
+ " <td>0.480769</td>\n",
+ " <td>0.574074</td>\n",
+ " <td>0.732314</td>\n",
+ " <td>0.736597</td>\n",
+ " <td>0.597983</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>36</td>\n",
+ " <td>0.025000</td>\n",
+ " <td>0.060461</td>\n",
+ " <td>0.589995</td>\n",
+ " <td>0.683121</td>\n",
+ " <td>0.262919</td>\n",
+ " <td>0.822199</td>\n",
+ " <td>0.517799</td>\n",
+ " <td>0.580645</td>\n",
+ " <td>0.736641</td>\n",
+ " <td>0.745312</td>\n",
+ " <td>0.617329</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>37</td>\n",
+ " <td>0.025000</td>\n",
+ " <td>0.064956</td>\n",
+ " <td>0.548047</td>\n",
+ " <td>0.685271</td>\n",
+ " <td>0.175471</td>\n",
+ " <td>0.827378</td>\n",
+ " <td>0.547771</td>\n",
+ " <td>0.618834</td>\n",
+ " <td>0.742968</td>\n",
+ " <td>0.736089</td>\n",
+ " <td>0.610229</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>38</td>\n",
+ " <td>0.025000</td>\n",
+ " <td>0.063804</td>\n",
+ " <td>0.539435</td>\n",
+ " <td>0.701887</td>\n",
+ " <td>0.270973</td>\n",
+ " <td>0.813606</td>\n",
+ " <td>0.541966</td>\n",
+ " <td>0.600000</td>\n",
+ " <td>0.745540</td>\n",
+ " <td>0.732331</td>\n",
+ " <td>0.618217</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>39</td>\n",
+ " <td>0.015200</td>\n",
+ " <td>0.067724</td>\n",
+ " <td>0.533308</td>\n",
+ " <td>0.642036</td>\n",
+ " <td>0.190905</td>\n",
+ " <td>0.821376</td>\n",
+ " <td>0.548896</td>\n",
+ " <td>0.600000</td>\n",
+ " <td>0.743961</td>\n",
+ " <td>0.718573</td>\n",
+ " <td>0.599882</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>40</td>\n",
+ " <td>0.015200</td>\n",
+ " <td>0.068417</td>\n",
+ " <td>0.533283</td>\n",
+ " <td>0.650718</td>\n",
+ " <td>0.182653</td>\n",
+ " <td>0.812395</td>\n",
+ " <td>0.601190</td>\n",
+ " <td>0.593607</td>\n",
+ " <td>0.744186</td>\n",
+ " <td>0.722561</td>\n",
+ " <td>0.605074</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>41</td>\n",
+ " <td>0.015200</td>\n",
+ " <td>0.068958</td>\n",
+ " <td>0.546763</td>\n",
+ " <td>0.678769</td>\n",
+ " <td>0.234627</td>\n",
+ " <td>0.772878</td>\n",
+ " <td>0.596859</td>\n",
+ " <td>0.593607</td>\n",
+ " <td>0.757374</td>\n",
+ " <td>0.730038</td>\n",
+ " <td>0.613864</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>42</td>\n",
+ " <td>0.015200</td>\n",
+ " <td>0.067570</td>\n",
+ " <td>0.550757</td>\n",
+ " <td>0.704247</td>\n",
+ " <td>0.213687</td>\n",
+ " <td>0.826460</td>\n",
+ " <td>0.561404</td>\n",
+ " <td>0.606335</td>\n",
+ " <td>0.748047</td>\n",
+ " <td>0.718676</td>\n",
+ " <td>0.616201</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>43</td>\n",
+ " <td>0.015200</td>\n",
+ " <td>0.069913</td>\n",
+ " <td>0.527344</td>\n",
+ " <td>0.673583</td>\n",
+ " <td>0.235346</td>\n",
+ " <td>0.813295</td>\n",
+ " <td>0.568966</td>\n",
+ " <td>0.593607</td>\n",
+ " <td>0.762808</td>\n",
+ " <td>0.707539</td>\n",
+ " <td>0.610311</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>44</td>\n",
+ " <td>0.015200</td>\n",
+ " <td>0.066919</td>\n",
+ " <td>0.548339</td>\n",
+ " <td>0.702537</td>\n",
+ " <td>0.250538</td>\n",
+ " <td>0.811425</td>\n",
+ " <td>0.594164</td>\n",
+ " <td>0.576744</td>\n",
+ " <td>0.759615</td>\n",
+ " <td>0.744470</td>\n",
+ " <td>0.623479</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>45</td>\n",
+ " <td>0.015200</td>\n",
+ " <td>0.070344</td>\n",
+ " <td>0.538619</td>\n",
+ " <td>0.684670</td>\n",
+ " <td>0.205105</td>\n",
+ " <td>0.812809</td>\n",
+ " <td>0.589474</td>\n",
+ " <td>0.589862</td>\n",
+ " <td>0.756964</td>\n",
+ " <td>0.735669</td>\n",
+ " <td>0.614147</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>46</td>\n",
+ " <td>0.015200</td>\n",
+ " <td>0.071540</td>\n",
+ " <td>0.509841</td>\n",
+ " <td>0.705255</td>\n",
+ " <td>0.239027</td>\n",
+ " <td>0.813519</td>\n",
+ " <td>0.552279</td>\n",
+ " <td>0.603604</td>\n",
+ " <td>0.756757</td>\n",
+ " <td>0.727416</td>\n",
+ " <td>0.613462</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>47</td>\n",
+ " <td>0.010700</td>\n",
+ " <td>0.070772</td>\n",
+ " <td>0.548339</td>\n",
+ " <td>0.693146</td>\n",
+ " <td>0.244704</td>\n",
+ " <td>0.818304</td>\n",
+ " <td>0.559767</td>\n",
+ " <td>0.587156</td>\n",
+ " <td>0.756757</td>\n",
+ " <td>0.723200</td>\n",
+ " <td>0.616422</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>48</td>\n",
+ " <td>0.010700</td>\n",
+ " <td>0.070882</td>\n",
+ " <td>0.547571</td>\n",
+ " <td>0.685490</td>\n",
+ " <td>0.221344</td>\n",
+ " <td>0.810738</td>\n",
+ " <td>0.541176</td>\n",
+ " <td>0.606335</td>\n",
+ " <td>0.762357</td>\n",
+ " <td>0.725745</td>\n",
+ " <td>0.612595</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>49</td>\n",
+ " <td>0.010700</td>\n",
+ " <td>0.070016</td>\n",
+ " <td>0.543567</td>\n",
+ " <td>0.702128</td>\n",
+ " <td>0.270270</td>\n",
+ " <td>0.810131</td>\n",
+ " <td>0.594059</td>\n",
+ " <td>0.606335</td>\n",
+ " <td>0.759924</td>\n",
+ " <td>0.748638</td>\n",
+ " <td>0.629382</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>50</td>\n",
+ " <td>0.010700</td>\n",
+ " <td>0.072628</td>\n",
+ " <td>0.542216</td>\n",
+ " <td>0.679186</td>\n",
+ " <td>0.249678</td>\n",
+ " <td>0.808700</td>\n",
+ " <td>0.567568</td>\n",
+ " <td>0.616071</td>\n",
+ " <td>0.763810</td>\n",
+ " <td>0.738609</td>\n",
+ " <td>0.620730</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>51</td>\n",
+ " <td>0.010700</td>\n",
+ " <td>0.071187</td>\n",
+ " <td>0.553411</td>\n",
+ " <td>0.674033</td>\n",
+ " <td>0.268354</td>\n",
+ " <td>0.805851</td>\n",
+ " <td>0.582888</td>\n",
+ " <td>0.600000</td>\n",
+ " <td>0.760994</td>\n",
+ " <td>0.726422</td>\n",
+ " <td>0.621494</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>52</td>\n",
+ " <td>0.010700</td>\n",
+ " <td>0.071963</td>\n",
+ " <td>0.550415</td>\n",
+ " <td>0.685848</td>\n",
+ " <td>0.259574</td>\n",
+ " <td>0.813526</td>\n",
+ " <td>0.593750</td>\n",
+ " <td>0.584475</td>\n",
+ " <td>0.757116</td>\n",
+ " <td>0.727129</td>\n",
+ " <td>0.621479</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>53</td>\n",
+ " <td>0.010700</td>\n",
+ " <td>0.071476</td>\n",
+ " <td>0.561372</td>\n",
+ " <td>0.694253</td>\n",
+ " <td>0.276007</td>\n",
+ " <td>0.809990</td>\n",
+ " <td>0.573727</td>\n",
+ " <td>0.609865</td>\n",
+ " <td>0.760994</td>\n",
+ " <td>0.745843</td>\n",
+ " <td>0.629006</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>54</td>\n",
+ " <td>0.008000</td>\n",
+ " <td>0.072140</td>\n",
+ " <td>0.555779</td>\n",
+ " <td>0.685848</td>\n",
+ " <td>0.263158</td>\n",
+ " <td>0.809853</td>\n",
+ " <td>0.584767</td>\n",
+ " <td>0.587156</td>\n",
+ " <td>0.759661</td>\n",
+ " <td>0.737405</td>\n",
+ " <td>0.622953</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>55</td>\n",
+ " <td>0.008000</td>\n",
+ " <td>0.071148</td>\n",
+ " <td>0.563626</td>\n",
+ " <td>0.693846</td>\n",
+ " <td>0.268644</td>\n",
+ " <td>0.812698</td>\n",
+ " <td>0.614583</td>\n",
+ " <td>0.669528</td>\n",
+ " <td>0.753946</td>\n",
+ " <td>0.736518</td>\n",
+ " <td>0.639174</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>56</td>\n",
+ " <td>0.008000</td>\n",
+ " <td>0.072347</td>\n",
+ " <td>0.563206</td>\n",
+ " <td>0.685490</td>\n",
+ " <td>0.278057</td>\n",
+ " <td>0.813593</td>\n",
+ " <td>0.597468</td>\n",
+ " <td>0.606335</td>\n",
+ " <td>0.753541</td>\n",
+ " <td>0.732778</td>\n",
+ " <td>0.628809</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>57</td>\n",
+ " <td>0.008000</td>\n",
+ " <td>0.072100</td>\n",
+ " <td>0.552941</td>\n",
+ " <td>0.684251</td>\n",
+ " <td>0.298728</td>\n",
+ " <td>0.821290</td>\n",
+ " <td>0.596859</td>\n",
+ " <td>0.612613</td>\n",
+ " <td>0.752363</td>\n",
+ " <td>0.731707</td>\n",
+ " <td>0.631344</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>58</td>\n",
+ " <td>0.008000</td>\n",
+ " <td>0.072843</td>\n",
+ " <td>0.549244</td>\n",
+ " <td>0.699237</td>\n",
+ " <td>0.290456</td>\n",
+ " <td>0.821850</td>\n",
+ " <td>0.588542</td>\n",
+ " <td>0.603604</td>\n",
+ " <td>0.751174</td>\n",
+ " <td>0.732103</td>\n",
+ " <td>0.629526</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>59</td>\n",
+ " <td>0.008000</td>\n",
+ " <td>0.073446</td>\n",
+ " <td>0.545723</td>\n",
+ " <td>0.706677</td>\n",
+ " <td>0.277801</td>\n",
+ " <td>0.808947</td>\n",
+ " <td>0.590674</td>\n",
+ " <td>0.606335</td>\n",
+ " <td>0.760037</td>\n",
+ " <td>0.731915</td>\n",
+ " <td>0.628514</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>60</td>\n",
+ " <td>0.008000</td>\n",
+ " <td>0.068292</td>\n",
+ " <td>0.571429</td>\n",
+ " <td>0.710546</td>\n",
+ " <td>0.323144</td>\n",
+ " <td>0.806643</td>\n",
+ " <td>0.625323</td>\n",
+ " <td>0.596330</td>\n",
+ " <td>0.753176</td>\n",
+ " <td>0.735294</td>\n",
+ " <td>0.640236</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>61</td>\n",
+ " <td>0.008000</td>\n",
+ " <td>0.069890</td>\n",
+ " <td>0.621654</td>\n",
+ " <td>0.701807</td>\n",
+ " <td>0.293851</td>\n",
+ " <td>0.789474</td>\n",
+ " <td>0.576819</td>\n",
+ " <td>0.666667</td>\n",
+ " <td>0.747331</td>\n",
+ " <td>0.627998</td>\n",
+ " <td>0.628200</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>62</td>\n",
+ " <td>0.006900</td>\n",
+ " <td>0.077804</td>\n",
+ " <td>0.556911</td>\n",
+ " <td>0.691603</td>\n",
+ " <td>0.256562</td>\n",
+ " <td>0.792987</td>\n",
+ " <td>0.570605</td>\n",
+ " <td>0.600000</td>\n",
+ " <td>0.774443</td>\n",
+ " <td>0.664879</td>\n",
+ " <td>0.613499</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>63</td>\n",
+ " <td>0.006900</td>\n",
+ " <td>0.075310</td>\n",
+ " <td>0.545927</td>\n",
+ " <td>0.676012</td>\n",
+ " <td>0.246670</td>\n",
+ " <td>0.813548</td>\n",
+ " <td>0.587601</td>\n",
+ " <td>0.602740</td>\n",
+ " <td>0.693416</td>\n",
+ " <td>0.682464</td>\n",
+ " <td>0.606047</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>64</td>\n",
+ " <td>0.006900</td>\n",
+ " <td>0.075709</td>\n",
+ " <td>0.567001</td>\n",
+ " <td>0.693215</td>\n",
+ " <td>0.254287</td>\n",
+ " <td>0.810577</td>\n",
+ " <td>0.600567</td>\n",
+ " <td>0.600000</td>\n",
+ " <td>0.779727</td>\n",
+ " <td>0.733489</td>\n",
+ " <td>0.629858</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>65</td>\n",
+ " <td>0.006900</td>\n",
+ " <td>0.074874</td>\n",
+ " <td>0.555064</td>\n",
+ " <td>0.700515</td>\n",
+ " <td>0.267559</td>\n",
+ " <td>0.807587</td>\n",
+ " <td>0.584856</td>\n",
+ " <td>0.600000</td>\n",
+ " <td>0.785579</td>\n",
+ " <td>0.719940</td>\n",
+ " <td>0.627638</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>66</td>\n",
+ " <td>0.006900</td>\n",
+ " <td>0.076945</td>\n",
+ " <td>0.555310</td>\n",
+ " <td>0.683230</td>\n",
+ " <td>0.259244</td>\n",
+ " <td>0.809295</td>\n",
+ " <td>0.587342</td>\n",
+ " <td>0.600000</td>\n",
+ " <td>0.777567</td>\n",
+ " <td>0.734819</td>\n",
+ " <td>0.625851</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>67</td>\n",
+ " <td>0.006900</td>\n",
+ " <td>0.079016</td>\n",
+ " <td>0.537566</td>\n",
+ " <td>0.693313</td>\n",
+ " <td>0.269021</td>\n",
+ " <td>0.809382</td>\n",
+ " <td>0.594458</td>\n",
+ " <td>0.600000</td>\n",
+ " <td>0.767677</td>\n",
+ " <td>0.735791</td>\n",
+ " <td>0.625901</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>68</td>\n",
+ " <td>0.006900</td>\n",
+ " <td>0.076534</td>\n",
+ " <td>0.545013</td>\n",
+ " <td>0.688752</td>\n",
+ " <td>0.255588</td>\n",
+ " <td>0.803952</td>\n",
+ " <td>0.614973</td>\n",
+ " <td>0.606335</td>\n",
+ " <td>0.759070</td>\n",
+ " <td>0.742642</td>\n",
+ " <td>0.627041</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>69</td>\n",
+ " <td>0.006900</td>\n",
+ " <td>0.079277</td>\n",
+ " <td>0.540419</td>\n",
+ " <td>0.705444</td>\n",
+ " <td>0.253413</td>\n",
+ " <td>0.804418</td>\n",
+ " <td>0.597884</td>\n",
+ " <td>0.634361</td>\n",
+ " <td>0.751456</td>\n",
+ " <td>0.732752</td>\n",
+ " <td>0.627518</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>70</td>\n",
+ " <td>0.006300</td>\n",
+ " <td>0.078259</td>\n",
+ " <td>0.547517</td>\n",
+ " <td>0.696510</td>\n",
+ " <td>0.283223</td>\n",
+ " <td>0.811007</td>\n",
+ " <td>0.579088</td>\n",
+ " <td>0.612613</td>\n",
+ " <td>0.753623</td>\n",
+ " <td>0.729430</td>\n",
+ " <td>0.626626</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>71</td>\n",
+ " <td>0.006300</td>\n",
+ " <td>0.078232</td>\n",
+ " <td>0.546977</td>\n",
+ " <td>0.703488</td>\n",
+ " <td>0.248394</td>\n",
+ " <td>0.807770</td>\n",
+ " <td>0.598425</td>\n",
+ " <td>0.649573</td>\n",
+ " <td>0.752381</td>\n",
+ " <td>0.737819</td>\n",
+ " <td>0.630603</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>72</td>\n",
+ " <td>0.006300</td>\n",
+ " <td>0.077648</td>\n",
+ " <td>0.554119</td>\n",
+ " <td>0.699552</td>\n",
+ " <td>0.256694</td>\n",
+ " <td>0.806284</td>\n",
+ " <td>0.598425</td>\n",
+ " <td>0.603604</td>\n",
+ " <td>0.753098</td>\n",
+ " <td>0.742378</td>\n",
+ " <td>0.626769</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>73</td>\n",
+ " <td>0.006300</td>\n",
+ " <td>0.077358</td>\n",
+ " <td>0.557540</td>\n",
+ " <td>0.701728</td>\n",
+ " <td>0.284053</td>\n",
+ " <td>0.805705</td>\n",
+ " <td>0.605598</td>\n",
+ " <td>0.607143</td>\n",
+ " <td>0.753976</td>\n",
+ " <td>0.737557</td>\n",
+ " <td>0.631662</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>74</td>\n",
+ " <td>0.006300</td>\n",
+ " <td>0.077561</td>\n",
+ " <td>0.564262</td>\n",
+ " <td>0.702176</td>\n",
+ " <td>0.278954</td>\n",
+ " <td>0.806734</td>\n",
+ " <td>0.591623</td>\n",
+ " <td>0.609865</td>\n",
+ " <td>0.757835</td>\n",
+ " <td>0.740061</td>\n",
+ " <td>0.631439</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>75</td>\n",
+ " <td>0.006300</td>\n",
+ " <td>0.078310</td>\n",
+ " <td>0.565003</td>\n",
+ " <td>0.701201</td>\n",
+ " <td>0.263357</td>\n",
+ " <td>0.803329</td>\n",
+ " <td>0.589474</td>\n",
+ " <td>0.600897</td>\n",
+ " <td>0.755471</td>\n",
+ " <td>0.742726</td>\n",
+ " <td>0.627682</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>76</td>\n",
+ " <td>0.006300</td>\n",
+ " <td>0.078413</td>\n",
+ " <td>0.567366</td>\n",
+ " <td>0.703148</td>\n",
+ " <td>0.270180</td>\n",
+ " <td>0.802114</td>\n",
+ " <td>0.576000</td>\n",
+ " <td>0.607143</td>\n",
+ " <td>0.757776</td>\n",
+ " <td>0.746544</td>\n",
+ " <td>0.628784</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>77</td>\n",
+ " <td>0.005000</td>\n",
+ " <td>0.078019</td>\n",
+ " <td>0.564796</td>\n",
+ " <td>0.705438</td>\n",
+ " <td>0.273219</td>\n",
+ " <td>0.805654</td>\n",
+ " <td>0.596306</td>\n",
+ " <td>0.616071</td>\n",
+ " <td>0.757974</td>\n",
+ " <td>0.740684</td>\n",
+ " <td>0.632518</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>78</td>\n",
+ " <td>0.005000</td>\n",
+ " <td>0.077703</td>\n",
+ " <td>0.566892</td>\n",
+ " <td>0.707646</td>\n",
+ " <td>0.275145</td>\n",
+ " <td>0.805477</td>\n",
+ " <td>0.587302</td>\n",
+ " <td>0.616071</td>\n",
+ " <td>0.756808</td>\n",
+ " <td>0.742857</td>\n",
+ " <td>0.632275</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>79</td>\n",
+ " <td>0.005000</td>\n",
+ " <td>0.078168</td>\n",
+ " <td>0.569659</td>\n",
+ " <td>0.704460</td>\n",
+ " <td>0.265828</td>\n",
+ " <td>0.806930</td>\n",
+ " <td>0.577540</td>\n",
+ " <td>0.603604</td>\n",
+ " <td>0.755005</td>\n",
+ " <td>0.750964</td>\n",
+ " <td>0.629249</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>80</td>\n",
+ " <td>0.005000</td>\n",
+ " <td>0.078695</td>\n",
+ " <td>0.568617</td>\n",
+ " <td>0.707391</td>\n",
+ " <td>0.269616</td>\n",
+ " <td>0.808101</td>\n",
+ " <td>0.579787</td>\n",
+ " <td>0.628319</td>\n",
+ " <td>0.755258</td>\n",
+ " <td>0.743570</td>\n",
+ " <td>0.632582</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>81</td>\n",
+ " <td>0.005000</td>\n",
+ " <td>0.079567</td>\n",
+ " <td>0.563530</td>\n",
+ " <td>0.704545</td>\n",
+ " <td>0.258365</td>\n",
+ " <td>0.808841</td>\n",
+ " <td>0.582011</td>\n",
+ " <td>0.622222</td>\n",
+ " <td>0.756705</td>\n",
+ " <td>0.743169</td>\n",
+ " <td>0.629924</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>82</td>\n",
+ " <td>0.005000</td>\n",
+ " <td>0.079116</td>\n",
+ " <td>0.566155</td>\n",
+ " <td>0.704012</td>\n",
+ " <td>0.264024</td>\n",
+ " <td>0.809779</td>\n",
+ " <td>0.579787</td>\n",
+ " <td>0.622222</td>\n",
+ " <td>0.760267</td>\n",
+ " <td>0.738413</td>\n",
+ " <td>0.630582</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>83</td>\n",
+ " <td>0.005000</td>\n",
+ " <td>0.079035</td>\n",
+ " <td>0.570076</td>\n",
+ " <td>0.703563</td>\n",
+ " <td>0.266835</td>\n",
+ " <td>0.808732</td>\n",
+ " <td>0.587302</td>\n",
+ " <td>0.634361</td>\n",
+ " <td>0.759542</td>\n",
+ " <td>0.739812</td>\n",
+ " <td>0.633778</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>84</td>\n",
+ " <td>0.005000</td>\n",
+ " <td>0.079356</td>\n",
+ " <td>0.568981</td>\n",
+ " <td>0.704545</td>\n",
+ " <td>0.266217</td>\n",
+ " <td>0.807873</td>\n",
+ " <td>0.587302</td>\n",
+ " <td>0.660944</td>\n",
+ " <td>0.761450</td>\n",
+ " <td>0.738824</td>\n",
+ " <td>0.637017</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>85</td>\n",
+ " <td>0.004500</td>\n",
+ " <td>0.079107</td>\n",
+ " <td>0.569973</td>\n",
+ " <td>0.707207</td>\n",
+ " <td>0.259619</td>\n",
+ " <td>0.807485</td>\n",
+ " <td>0.591623</td>\n",
+ " <td>0.655172</td>\n",
+ " <td>0.764212</td>\n",
+ " <td>0.740566</td>\n",
+ " <td>0.636982</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>86</td>\n",
+ " <td>0.004500</td>\n",
+ " <td>0.079421</td>\n",
+ " <td>0.568827</td>\n",
+ " <td>0.705971</td>\n",
+ " <td>0.258694</td>\n",
+ " <td>0.809644</td>\n",
+ " <td>0.585752</td>\n",
+ " <td>0.652361</td>\n",
+ " <td>0.765363</td>\n",
+ " <td>0.737743</td>\n",
+ " <td>0.635544</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>87</td>\n",
+ " <td>0.004500</td>\n",
+ " <td>0.079341</td>\n",
+ " <td>0.567676</td>\n",
+ " <td>0.707554</td>\n",
+ " <td>0.257106</td>\n",
+ " <td>0.811289</td>\n",
+ " <td>0.578947</td>\n",
+ " <td>0.649351</td>\n",
+ " <td>0.762984</td>\n",
+ " <td>0.737984</td>\n",
+ " <td>0.634112</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>88</td>\n",
+ " <td>0.004500</td>\n",
+ " <td>0.079288</td>\n",
+ " <td>0.567780</td>\n",
+ " <td>0.706497</td>\n",
+ " <td>0.261311</td>\n",
+ " <td>0.811236</td>\n",
+ " <td>0.582677</td>\n",
+ " <td>0.646552</td>\n",
+ " <td>0.764428</td>\n",
+ " <td>0.739705</td>\n",
+ " <td>0.635023</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>89</td>\n",
+ " <td>0.004500</td>\n",
+ " <td>0.079334</td>\n",
+ " <td>0.567676</td>\n",
+ " <td>0.708520</td>\n",
+ " <td>0.270475</td>\n",
+ " <td>0.811135</td>\n",
+ " <td>0.579634</td>\n",
+ " <td>0.646552</td>\n",
+ " <td>0.762085</td>\n",
+ " <td>0.739130</td>\n",
+ " <td>0.635651</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>90</td>\n",
+ " <td>0.004500</td>\n",
+ " <td>0.079473</td>\n",
+ " <td>0.567153</td>\n",
+ " <td>0.708520</td>\n",
+ " <td>0.267452</td>\n",
+ " <td>0.810241</td>\n",
+ " <td>0.582677</td>\n",
+ " <td>0.652361</td>\n",
+ " <td>0.763981</td>\n",
+ " <td>0.738962</td>\n",
+ " <td>0.636418</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>91</td>\n",
+ " <td>0.004500</td>\n",
+ " <td>0.079833</td>\n",
+ " <td>0.567153</td>\n",
+ " <td>0.710645</td>\n",
+ " <td>0.264247</td>\n",
+ " <td>0.811074</td>\n",
+ " <td>0.580475</td>\n",
+ " <td>0.652361</td>\n",
+ " <td>0.763258</td>\n",
+ " <td>0.737984</td>\n",
+ " <td>0.635899</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>92</td>\n",
+ " <td>0.004500</td>\n",
+ " <td>0.080144</td>\n",
+ " <td>0.567937</td>\n",
+ " <td>0.709434</td>\n",
+ " <td>0.260722</td>\n",
+ " <td>0.810212</td>\n",
+ " <td>0.576720</td>\n",
+ " <td>0.652361</td>\n",
+ " <td>0.764706</td>\n",
+ " <td>0.739130</td>\n",
+ " <td>0.635153</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>93</td>\n",
+ " <td>0.004200</td>\n",
+ " <td>0.080148</td>\n",
+ " <td>0.567050</td>\n",
+ " <td>0.706505</td>\n",
+ " <td>0.261349</td>\n",
+ " <td>0.810666</td>\n",
+ " <td>0.576720</td>\n",
+ " <td>0.652361</td>\n",
+ " <td>0.764259</td>\n",
+ " <td>0.740452</td>\n",
+ " <td>0.634920</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>94</td>\n",
+ " <td>0.004200</td>\n",
+ " <td>0.080076</td>\n",
+ " <td>0.567202</td>\n",
+ " <td>0.704012</td>\n",
+ " <td>0.263625</td>\n",
+ " <td>0.811107</td>\n",
+ " <td>0.576720</td>\n",
+ " <td>0.643478</td>\n",
+ " <td>0.763533</td>\n",
+ " <td>0.737743</td>\n",
+ " <td>0.633428</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>95</td>\n",
+ " <td>0.004200</td>\n",
+ " <td>0.080163</td>\n",
+ " <td>0.568563</td>\n",
+ " <td>0.702580</td>\n",
+ " <td>0.260465</td>\n",
+ " <td>0.810775</td>\n",
+ " <td>0.585752</td>\n",
+ " <td>0.631579</td>\n",
+ " <td>0.762808</td>\n",
+ " <td>0.737500</td>\n",
+ " <td>0.632503</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>96</td>\n",
+ " <td>0.004200</td>\n",
+ " <td>0.080179</td>\n",
+ " <td>0.568931</td>\n",
+ " <td>0.703196</td>\n",
+ " <td>0.259949</td>\n",
+ " <td>0.812063</td>\n",
+ " <td>0.583554</td>\n",
+ " <td>0.625551</td>\n",
+ " <td>0.762808</td>\n",
+ " <td>0.736513</td>\n",
+ " <td>0.631571</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>97</td>\n",
+ " <td>0.004200</td>\n",
+ " <td>0.080223</td>\n",
+ " <td>0.568774</td>\n",
+ " <td>0.703196</td>\n",
+ " <td>0.260575</td>\n",
+ " <td>0.812604</td>\n",
+ " <td>0.587302</td>\n",
+ " <td>0.631579</td>\n",
+ " <td>0.763981</td>\n",
+ " <td>0.737500</td>\n",
+ " <td>0.633189</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>98</td>\n",
+ " <td>0.004200</td>\n",
+ " <td>0.080253</td>\n",
+ " <td>0.568513</td>\n",
+ " <td>0.702662</td>\n",
+ " <td>0.261935</td>\n",
+ " <td>0.812875</td>\n",
+ " <td>0.591029</td>\n",
+ " <td>0.631579</td>\n",
+ " <td>0.762535</td>\n",
+ " <td>0.738486</td>\n",
+ " <td>0.633702</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>99</td>\n",
+ " <td>0.004200</td>\n",
+ " <td>0.080378</td>\n",
+ " <td>0.567834</td>\n",
+ " <td>0.702128</td>\n",
+ " <td>0.261200</td>\n",
+ " <td>0.812427</td>\n",
+ " <td>0.585752</td>\n",
+ " <td>0.631579</td>\n",
+ " <td>0.763705</td>\n",
+ " <td>0.739062</td>\n",
+ " <td>0.632961</td>\n",
+ " </tr>\n",
+ " <tr>\n",
+ " <td>100</td>\n",
+ " <td>0.004000</td>\n",
+ " <td>0.080378</td>\n",
+ " <td>0.567834</td>\n",
+ " <td>0.703113</td>\n",
+ " <td>0.261200</td>\n",
+ " <td>0.812563</td>\n",
+ " <td>0.585752</td>\n",
+ " <td>0.631579</td>\n",
+ " <td>0.763705</td>\n",
+ " <td>0.739062</td>\n",
+ " <td>0.633101</td>\n",
+ " </tr>\n",
+ " </tbody>\n",
+ "</table><p>"
+ ],
+ "text/plain": [
+ "<IPython.core.display.HTML object>"
+ ]
+ },
+ "metadata": {},
+ "output_type": "display_data"
+ },
+ {
+ "data": {
+ "text/plain": [
+ "TrainOutput(global_step=1300, training_loss=0.03139937967061997, metrics={'train_runtime': 548.1168, 'train_samples_per_second': 37.583, 'train_steps_per_second': 2.372, 'total_flos': 2691526921113600.0, 'train_loss': 0.03139937967061997, 'epoch': 100.0})"
+ ]
+ },
+ "execution_count": 13,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "# Initialize the callback\n",
+ "metrics_callback = TrainingMetricsCallback()\n",
+ "\n",
+ "def model_init():\n",
+ " # For reproducibility\n",
+ " return RobertaForSpanCategorization.from_pretrained(modelId, id2label=id2label, label2id=label2id)\n",
+ "\n",
+ "trainer = Trainer(\n",
+ " model_init=model_init,\n",
+ " args=training_args,\n",
+ " train_dataset=tokenized_train_ds,\n",
+ " eval_dataset=tokenized_val_ds,\n",
+ " data_collator=data_collator,\n",
+ " tokenizer=tokenizer,\n",
+ " compute_metrics=compute_metrics,\n",
+ " callbacks=[metrics_callback]\n",
+ ")\n",
+ "trainer.train()"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 21,
+ "id": "6b029310-5257-41d8-a5f7-3bbb2021ebae",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "image/png": "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",
+ "text/plain": [
+ "<Figure size 1000x500 with 1 Axes>"
+ ]
+ },
+ "metadata": {},
+ "output_type": "display_data"
+ }
+ ],
+ "source": [
+ "# Plot the training and evaluation metrics\n",
+ "df = pd.DataFrame({'step': metrics_callback.steps})\n",
+ "df['macro_f1'] = pd.Series(metrics_callback.macro_f1)\n",
+ "\n",
+ "plt.figure(figsize=(10, 5))\n",
+ "plt.plot(df['step'], df['macro_f1'], label='Macro F1')\n",
+ "plt.xlabel('Step')\n",
+ "plt.ylabel('Metrics')\n",
+ "plt.legend()\n",
+ "plt.title('Training and Evaluation Metrics over Steps')\n",
+ "\n",
+ "plt.savefig('../../results/BERT/legal_bert.png')\n",
+ "plt.show()"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 15,
+ "id": "89c997e6a944bc70",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "trainer.model.save_pretrained(\"../../models/Fine-tuned_LegalCamemBERT-base\")"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "05f2cf51-2a76-4de6-9b22-af7dda1eb805",
+ "metadata": {},
+ "source": [
+ "# Inference"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 16,
+ "id": "e48a67b7-814d-46d5-a079-fafc5b6adf86",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "model = RobertaForSpanCategorization.from_pretrained(\"../../models/Fine-tuned_LegalCamemBERT-base\")\n",
+ "tokenizer = AutoTokenizer.from_pretrained(modelId)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 17,
+ "id": "4cb75af5-234f-4863-94f8-45a364e1b877",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "def get_offsets_and_predicted_tags(example: str, model, tokenizer, threshold=0):\n",
+ " \"\"\"\n",
+ " Get prediction of model on example, using tokenizer\n",
+ " Args:\n",
+ " - example (str): The input text\n",
+ " - model: The span categorizer\n",
+ " - tokenizer: The tokenizer\n",
+ " - threshold: The threshold to decide whether the token should belong to the label. Default to 0, which corresponds to probability 0.5.\n",
+ " Returns:\n",
+ " - List of (token, tags, offset) for each token.\n",
+ " \"\"\"\n",
+ " # Tokenize the sentence to retrieve the tokens and offset mappings\n",
+ " raw_encoded_example = tokenizer(example, return_offsets_mapping=True)\n",
+ " encoded_example = tokenizer(example, return_tensors=\"pt\")\n",
+ " \n",
+ " # Call the model. The output LxK-tensor where L is the number of tokens, K is the number of classes\n",
+ " out = model(**encoded_example)[\"logits\"][0]\n",
+ " \n",
+ " # We assign to each token the classes whose logit is positive\n",
+ " predicted_tags = [[i for i, l in enumerate(logit) if l > threshold] for logit in out]\n",
+ " \n",
+ " return [{\"token\": token, \"tags\": tag, \"offset\": offset} for (token, tag, offset) \n",
+ " in zip(tokenizer.batch_decode(raw_encoded_example[\"input_ids\"]), \n",
+ " predicted_tags, \n",
+ " raw_encoded_example[\"offset_mapping\"])]"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 18,
+ "id": "314d7a28-ef8b-4367-8e6d-79a55951fcec",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "<s> - []\n",
+ "afin - []\n",
+ "de - []\n",
+ "vérifier - []\n",
+ "le - []\n",
+ "kilométrage - []\n",
+ " - []\n",
+ ", - []\n",
+ "pour - []\n",
+ "les - [3]\n",
+ "véhicules - [3]\n",
+ "équipés - [4]\n",
+ "d - [4]\n",
+ " - [4]\n",
+ "' - [4]\n",
+ "un - [4]\n",
+ "compteur - [4]\n",
+ "kilo - [4]\n",
+ "métrique - [4]\n",
+ " - []\n",
+ ", - []\n",
+ "les - [3]\n",
+ "informations - [3]\n",
+ "communiquées - [4]\n",
+ "lors - [4, 8]\n",
+ "du - [8]\n",
+ "précédent - [8]\n",
+ "contrôle - [8]\n",
+ "technique - [8]\n",
+ "sont - [1]\n",
+ "mises - [1]\n",
+ "à - [1]\n",
+ "la - [1]\n",
+ "disposition - [1]\n",
+ "des - [1]\n",
+ "organismes - [1, 2]\n",
+ "de - [1, 2]\n",
+ "contrôle - [1, 2]\n",
+ "technique - [1, 2]\n",
+ "dès - [4]\n",
+ "qu - [4]\n",
+ " - [4]\n",
+ "' - [4]\n",
+ "elles - [4]\n",
+ "sont - [4]\n",
+ "disponibles - [4]\n",
+ "par - [4]\n",
+ "voie - [4]\n",
+ "électronique - [4]\n",
+ " - []\n",
+ ". - []\n",
+ "</s> - []\n"
+ ]
+ }
+ ],
+ "source": [
+ "example = \"afin de vérifier le kilométrage , pour les véhicules équipés d ' un compteur kilométrique , les informations communiquées lors du précédent contrôle technique sont mises à la disposition des organismes de contrôle technique dès qu ' elles sont disponibles par voie électronique .\"\n",
+ "for item in get_offsets_and_predicted_tags(example, model, tokenizer):\n",
+ " print(f\"\"\"{item[\"token\"]:15} - {item[\"tags\"]}\"\"\")"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 19,
+ "id": "b0164373-ee70-4d63-a0c2-ac2197b37b22",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "afin de vérifier le kilométrage , pour les véhicules équipés d ' un compteur kilométrique , les informations communiquées lors du précédent contrôle technique sont mises à la disposition des organismes de contrôle technique dès qu ' elles sont disponibles par voie électronique .\n"
+ ]
+ },
+ {
+ "data": {
+ "text/plain": [
+ "[{'start': 39, 'end': 52, 'tag': 'artifact', 'text': 'les véhicules'},\n",
+ " {'start': 53,\n",
+ " 'end': 89,\n",
+ " 'tag': 'condition',\n",
+ " 'text': \"équipés d ' un compteur kilométrique\"},\n",
+ " {'start': 92, 'end': 108, 'tag': 'artifact', 'text': 'les informations'},\n",
+ " {'start': 109, 'end': 126, 'tag': 'condition', 'text': 'communiquées lors'},\n",
+ " {'start': 122,\n",
+ " 'end': 158,\n",
+ " 'tag': 'time',\n",
+ " 'text': 'lors du précédent contrôle technique'},\n",
+ " {'start': 159,\n",
+ " 'end': 223,\n",
+ " 'tag': 'action',\n",
+ " 'text': 'sont mises à la disposition des organismes de contrôle technique'},\n",
+ " {'start': 191,\n",
+ " 'end': 223,\n",
+ " 'tag': 'actor',\n",
+ " 'text': 'organismes de contrôle technique'},\n",
+ " {'start': 224,\n",
+ " 'end': 277,\n",
+ " 'tag': 'condition',\n",
+ " 'text': \"dès qu ' elles sont disponibles par voie électronique\"}]"
+ ]
+ },
+ "execution_count": 19,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "def get_tagged_groups(example: str, model, tokenizer):\n",
+ " \"\"\"\n",
+ " Get prediction of model on example, using tokenizer\n",
+ " Returns:\n",
+ " - List of spans under offset format {\"start\": ..., \"end\": ..., \"tag\": ...}, sorted by start, end then tag.\n",
+ " \"\"\"\n",
+ " offsets_and_tags = get_offsets_and_predicted_tags(example, model, tokenizer)\n",
+ " predicted_offsets = {l: [] for l in tag2id}\n",
+ " last_token_tags = []\n",
+ " for item in offsets_and_tags:\n",
+ " (start, end), tags = item[\"offset\"], item[\"tags\"]\n",
+ " \n",
+ " for label_id in tags:\n",
+ " tag = id2label[label_id]\n",
+ " if label_id not in last_token_tags and label2id[f\"{tag}\"] not in last_token_tags:\n",
+ " predicted_offsets[tag].append({\"start\": start, \"end\": end})\n",
+ " else:\n",
+ " predicted_offsets[tag][-1][\"end\"] = end\n",
+ " \n",
+ " last_token_tags = tags\n",
+ " \n",
+ " flatten_predicted_offsets = [{**v, \"tag\": k, \"text\": example[v[\"start\"]:v[\"end\"]]} \n",
+ " for k, v_list in predicted_offsets.items() for v in v_list if v[\"end\"] - v[\"start\"] >= 3]\n",
+ " flatten_predicted_offsets = sorted(flatten_predicted_offsets, \n",
+ " key = lambda row: (row[\"start\"], row[\"end\"], row[\"tag\"]))\n",
+ " return flatten_predicted_offsets\n",
+ "\n",
+ "print(example)\n",
+ "get_tagged_groups(example, model, tokenizer)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "id": "54213be5-8221-40a2-ac60-04670d83babc",
+ "metadata": {},
+ "outputs": [],
+ "source": []
+ }
+ ],
+ "metadata": {
+ "kernelspec": {
+ "display_name": "Python 3 (ipykernel)",
+ "language": "python",
+ "name": "python3"
+ },
+ "language_info": {
+ "codemirror_mode": {
+ "name": "ipython",
+ "version": 3
+ },
+ "file_extension": ".py",
+ "mimetype": "text/x-python",
+ "name": "python",
+ "nbconvert_exporter": "python",
+ "pygments_lexer": "ipython3",
+ "version": "3.11.0"
+ }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}
diff --git a/modules/camembert/camembert-classifier.ipynb b/modules/camembert/camembert-classifier.ipynb
deleted file mode 100644
index 83785f3..0000000
--- a/modules/camembert/camembert-classifier.ipynb
+++ /dev/null
@@ -1,662 +0,0 @@
-{
- "cells": [
- {
- "cell_type": "markdown",
- "id": "8ced2e3ca31fb46c",
- "metadata": {},
- "source": [
- "# Dataset"
- ]
- },
- {
- "cell_type": "code",
- "id": "757a8bf026156e77",
- "metadata": {
- "ExecuteTime": {
- "end_time": "2024-06-27T14:45:25.410725Z",
- "start_time": "2024-06-27T14:45:25.404357Z"
- }
- },
- "source": [
- "tag2id = {'action': 1, 'actor': 2, 'artifact': 3, 'condition': 4, 'location': 5, 'modality': 6, 'reference': 7, 'time': 8}\n",
- "id2tag = {v:k for k, v in tag2id.items()}"
- ],
- "outputs": [],
- "execution_count": 1
- },
- {
- "cell_type": "code",
- "id": "be3a4c320f9d4a5",
- "metadata": {
- "editable": true,
- "slideshow": {
- "slide_type": ""
- },
- "tags": [],
- "ExecuteTime": {
- "end_time": "2024-06-27T14:45:35.889919Z",
- "start_time": "2024-06-27T14:45:35.885841Z"
- }
- },
- "source": [
- "label2id = {\n",
- " 'O': 0,\n",
- " **{f'B-{k}': 2*v - 1 for k, v in tag2id.items()},\n",
- " **{f'I-{k}': 2*v for k, v in tag2id.items()}\n",
- "}\n",
- "\n",
- "id2label = {v:k for k, v in label2id.items()}"
- ],
- "outputs": [],
- "execution_count": 2
- },
- {
- "cell_type": "code",
- "id": "2aa2fefac95e7f04",
- "metadata": {
- "ExecuteTime": {
- "end_time": "2024-06-27T14:45:46.507242Z",
- "start_time": "2024-06-27T14:45:38.632569Z"
- }
- },
- "source": [
- "from datasets import Dataset\n",
- "train_ds = Dataset.from_json(\"../../data/annotations.train.jsonlines\")\n",
- "val_ds = Dataset.from_json(\"../../data/annotations.eval.jsonlines\")"
- ],
- "outputs": [
- {
- "data": {
- "text/plain": [
- "Generating train split: 0 examples [00:00, ? examples/s]"
- ],
- "application/vnd.jupyter.widget-view+json": {
- "version_major": 2,
- "version_minor": 0,
- "model_id": "c0f11a82572440d9bd1405c2e6ea6d2a"
- }
- },
- "metadata": {},
- "output_type": "display_data"
- },
- {
- "data": {
- "text/plain": [
- "Generating train split: 0 examples [00:00, ? examples/s]"
- ],
- "application/vnd.jupyter.widget-view+json": {
- "version_major": 2,
- "version_minor": 0,
- "model_id": "08366b1257d34fcd97ace0017fd3b395"
- }
- },
- "metadata": {},
- "output_type": "display_data"
- }
- ],
- "execution_count": 3
- },
- {
- "cell_type": "code",
- "id": "9e0a21356e7701a1",
- "metadata": {
- "ExecuteTime": {
- "end_time": "2024-06-27T14:45:52.188551Z",
- "start_time": "2024-06-27T14:45:52.185648Z"
- }
- },
- "source": [
- "modelId = '../../models/CamemBERT-base'"
- ],
- "outputs": [],
- "execution_count": 4
- },
- {
- "cell_type": "markdown",
- "id": "66e00d5a79a66753",
- "metadata": {},
- "source": [
- "# Tokenization"
- ]
- },
- {
- "cell_type": "code",
- "id": "e6459259f5ab2d98",
- "metadata": {
- "ExecuteTime": {
- "end_time": "2024-06-27T14:46:02.189231Z",
- "start_time": "2024-06-27T14:45:56.737350Z"
- }
- },
- "source": [
- "from transformers import AutoTokenizer\n",
- "tokenizer = AutoTokenizer.from_pretrained(modelId)"
- ],
- "outputs": [
- {
- "ename": "OSError",
- "evalue": "../../models/CamemBERT-base does not appear to have a file named config.json. Checkout 'https://huggingface.co/../../models/CamemBERT-base/None' for available files.",
- "output_type": "error",
- "traceback": [
- "\u001B[1;31m---------------------------------------------------------------------------\u001B[0m",
- "\u001B[1;31mOSError\u001B[0m Traceback (most recent call last)",
- "Cell \u001B[1;32mIn[5], line 2\u001B[0m\n\u001B[0;32m 1\u001B[0m \u001B[38;5;28;01mfrom\u001B[39;00m \u001B[38;5;21;01mtransformers\u001B[39;00m \u001B[38;5;28;01mimport\u001B[39;00m AutoTokenizer\n\u001B[1;32m----> 2\u001B[0m tokenizer \u001B[38;5;241m=\u001B[39m \u001B[43mAutoTokenizer\u001B[49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43mfrom_pretrained\u001B[49m\u001B[43m(\u001B[49m\u001B[43mmodelId\u001B[49m\u001B[43m)\u001B[49m\n",
- "File \u001B[1;32m~\\Desktop\\Sync\\Berger-Levrault\\Codes\\Ala\\Legal Concepts Extraction\\venv\\Lib\\site-packages\\transformers\\models\\auto\\tokenization_auto.py:773\u001B[0m, in \u001B[0;36mAutoTokenizer.from_pretrained\u001B[1;34m(cls, pretrained_model_name_or_path, *inputs, **kwargs)\u001B[0m\n\u001B[0;32m 771\u001B[0m \u001B[38;5;28;01mif\u001B[39;00m config_tokenizer_class \u001B[38;5;129;01mis\u001B[39;00m \u001B[38;5;28;01mNone\u001B[39;00m:\n\u001B[0;32m 772\u001B[0m \u001B[38;5;28;01mif\u001B[39;00m \u001B[38;5;129;01mnot\u001B[39;00m \u001B[38;5;28misinstance\u001B[39m(config, PretrainedConfig):\n\u001B[1;32m--> 773\u001B[0m config \u001B[38;5;241m=\u001B[39m \u001B[43mAutoConfig\u001B[49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43mfrom_pretrained\u001B[49m\u001B[43m(\u001B[49m\n\u001B[0;32m 774\u001B[0m \u001B[43m \u001B[49m\u001B[43mpretrained_model_name_or_path\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43mtrust_remote_code\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43mtrust_remote_code\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[43mkwargs\u001B[49m\n\u001B[0;32m 775\u001B[0m \u001B[43m \u001B[49m\u001B[43m)\u001B[49m\n\u001B[0;32m 776\u001B[0m config_tokenizer_class \u001B[38;5;241m=\u001B[39m config\u001B[38;5;241m.\u001B[39mtokenizer_class\n\u001B[0;32m 777\u001B[0m \u001B[38;5;28;01mif\u001B[39;00m \u001B[38;5;28mhasattr\u001B[39m(config, \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mauto_map\u001B[39m\u001B[38;5;124m\"\u001B[39m) \u001B[38;5;129;01mand\u001B[39;00m \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mAutoTokenizer\u001B[39m\u001B[38;5;124m\"\u001B[39m \u001B[38;5;129;01min\u001B[39;00m config\u001B[38;5;241m.\u001B[39mauto_map:\n",
- "File \u001B[1;32m~\\Desktop\\Sync\\Berger-Levrault\\Codes\\Ala\\Legal Concepts Extraction\\venv\\Lib\\site-packages\\transformers\\models\\auto\\configuration_auto.py:1100\u001B[0m, in \u001B[0;36mAutoConfig.from_pretrained\u001B[1;34m(cls, pretrained_model_name_or_path, **kwargs)\u001B[0m\n\u001B[0;32m 1097\u001B[0m trust_remote_code \u001B[38;5;241m=\u001B[39m kwargs\u001B[38;5;241m.\u001B[39mpop(\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mtrust_remote_code\u001B[39m\u001B[38;5;124m\"\u001B[39m, \u001B[38;5;28;01mNone\u001B[39;00m)\n\u001B[0;32m 1098\u001B[0m code_revision \u001B[38;5;241m=\u001B[39m kwargs\u001B[38;5;241m.\u001B[39mpop(\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mcode_revision\u001B[39m\u001B[38;5;124m\"\u001B[39m, \u001B[38;5;28;01mNone\u001B[39;00m)\n\u001B[1;32m-> 1100\u001B[0m config_dict, unused_kwargs \u001B[38;5;241m=\u001B[39m \u001B[43mPretrainedConfig\u001B[49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43mget_config_dict\u001B[49m\u001B[43m(\u001B[49m\u001B[43mpretrained_model_name_or_path\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[43mkwargs\u001B[49m\u001B[43m)\u001B[49m\n\u001B[0;32m 1101\u001B[0m has_remote_code \u001B[38;5;241m=\u001B[39m \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mauto_map\u001B[39m\u001B[38;5;124m\"\u001B[39m \u001B[38;5;129;01min\u001B[39;00m config_dict \u001B[38;5;129;01mand\u001B[39;00m \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mAutoConfig\u001B[39m\u001B[38;5;124m\"\u001B[39m \u001B[38;5;129;01min\u001B[39;00m config_dict[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mauto_map\u001B[39m\u001B[38;5;124m\"\u001B[39m]\n\u001B[0;32m 1102\u001B[0m has_local_code \u001B[38;5;241m=\u001B[39m \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mmodel_type\u001B[39m\u001B[38;5;124m\"\u001B[39m \u001B[38;5;129;01min\u001B[39;00m config_dict \u001B[38;5;129;01mand\u001B[39;00m config_dict[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mmodel_type\u001B[39m\u001B[38;5;124m\"\u001B[39m] \u001B[38;5;129;01min\u001B[39;00m CONFIG_MAPPING\n",
- "File \u001B[1;32m~\\Desktop\\Sync\\Berger-Levrault\\Codes\\Ala\\Legal Concepts Extraction\\venv\\Lib\\site-packages\\transformers\\configuration_utils.py:634\u001B[0m, in \u001B[0;36mPretrainedConfig.get_config_dict\u001B[1;34m(cls, pretrained_model_name_or_path, **kwargs)\u001B[0m\n\u001B[0;32m 632\u001B[0m original_kwargs \u001B[38;5;241m=\u001B[39m copy\u001B[38;5;241m.\u001B[39mdeepcopy(kwargs)\n\u001B[0;32m 633\u001B[0m \u001B[38;5;66;03m# Get config dict associated with the base config file\u001B[39;00m\n\u001B[1;32m--> 634\u001B[0m config_dict, kwargs \u001B[38;5;241m=\u001B[39m \u001B[38;5;28;43mcls\u001B[39;49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43m_get_config_dict\u001B[49m\u001B[43m(\u001B[49m\u001B[43mpretrained_model_name_or_path\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[43mkwargs\u001B[49m\u001B[43m)\u001B[49m\n\u001B[0;32m 635\u001B[0m \u001B[38;5;28;01mif\u001B[39;00m \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124m_commit_hash\u001B[39m\u001B[38;5;124m\"\u001B[39m \u001B[38;5;129;01min\u001B[39;00m config_dict:\n\u001B[0;32m 636\u001B[0m original_kwargs[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124m_commit_hash\u001B[39m\u001B[38;5;124m\"\u001B[39m] \u001B[38;5;241m=\u001B[39m config_dict[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124m_commit_hash\u001B[39m\u001B[38;5;124m\"\u001B[39m]\n",
- "File \u001B[1;32m~\\Desktop\\Sync\\Berger-Levrault\\Codes\\Ala\\Legal Concepts Extraction\\venv\\Lib\\site-packages\\transformers\\configuration_utils.py:689\u001B[0m, in \u001B[0;36mPretrainedConfig._get_config_dict\u001B[1;34m(cls, pretrained_model_name_or_path, **kwargs)\u001B[0m\n\u001B[0;32m 685\u001B[0m configuration_file \u001B[38;5;241m=\u001B[39m kwargs\u001B[38;5;241m.\u001B[39mpop(\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124m_configuration_file\u001B[39m\u001B[38;5;124m\"\u001B[39m, CONFIG_NAME)\n\u001B[0;32m 687\u001B[0m \u001B[38;5;28;01mtry\u001B[39;00m:\n\u001B[0;32m 688\u001B[0m \u001B[38;5;66;03m# Load from local folder or from cache or download from model Hub and cache\u001B[39;00m\n\u001B[1;32m--> 689\u001B[0m resolved_config_file \u001B[38;5;241m=\u001B[39m \u001B[43mcached_file\u001B[49m\u001B[43m(\u001B[49m\n\u001B[0;32m 690\u001B[0m \u001B[43m \u001B[49m\u001B[43mpretrained_model_name_or_path\u001B[49m\u001B[43m,\u001B[49m\n\u001B[0;32m 691\u001B[0m \u001B[43m \u001B[49m\u001B[43mconfiguration_file\u001B[49m\u001B[43m,\u001B[49m\n\u001B[0;32m 692\u001B[0m \u001B[43m \u001B[49m\u001B[43mcache_dir\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43mcache_dir\u001B[49m\u001B[43m,\u001B[49m\n\u001B[0;32m 693\u001B[0m \u001B[43m \u001B[49m\u001B[43mforce_download\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43mforce_download\u001B[49m\u001B[43m,\u001B[49m\n\u001B[0;32m 694\u001B[0m \u001B[43m \u001B[49m\u001B[43mproxies\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43mproxies\u001B[49m\u001B[43m,\u001B[49m\n\u001B[0;32m 695\u001B[0m \u001B[43m \u001B[49m\u001B[43mresume_download\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43mresume_download\u001B[49m\u001B[43m,\u001B[49m\n\u001B[0;32m 696\u001B[0m \u001B[43m \u001B[49m\u001B[43mlocal_files_only\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43mlocal_files_only\u001B[49m\u001B[43m,\u001B[49m\n\u001B[0;32m 697\u001B[0m \u001B[43m \u001B[49m\u001B[43mtoken\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43mtoken\u001B[49m\u001B[43m,\u001B[49m\n\u001B[0;32m 698\u001B[0m \u001B[43m \u001B[49m\u001B[43muser_agent\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43muser_agent\u001B[49m\u001B[43m,\u001B[49m\n\u001B[0;32m 699\u001B[0m \u001B[43m \u001B[49m\u001B[43mrevision\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43mrevision\u001B[49m\u001B[43m,\u001B[49m\n\u001B[0;32m 700\u001B[0m \u001B[43m \u001B[49m\u001B[43msubfolder\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43msubfolder\u001B[49m\u001B[43m,\u001B[49m\n\u001B[0;32m 701\u001B[0m \u001B[43m \u001B[49m\u001B[43m_commit_hash\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43mcommit_hash\u001B[49m\u001B[43m,\u001B[49m\n\u001B[0;32m 702\u001B[0m \u001B[43m \u001B[49m\u001B[43m)\u001B[49m\n\u001B[0;32m 703\u001B[0m commit_hash \u001B[38;5;241m=\u001B[39m extract_commit_hash(resolved_config_file, commit_hash)\n\u001B[0;32m 704\u001B[0m \u001B[38;5;28;01mexcept\u001B[39;00m \u001B[38;5;167;01mEnvironmentError\u001B[39;00m:\n\u001B[0;32m 705\u001B[0m \u001B[38;5;66;03m# Raise any environment error raise by `cached_file`. It will have a helpful error message adapted to\u001B[39;00m\n\u001B[0;32m 706\u001B[0m \u001B[38;5;66;03m# the original exception.\u001B[39;00m\n",
- "File \u001B[1;32m~\\Desktop\\Sync\\Berger-Levrault\\Codes\\Ala\\Legal Concepts Extraction\\venv\\Lib\\site-packages\\transformers\\utils\\hub.py:356\u001B[0m, in \u001B[0;36mcached_file\u001B[1;34m(path_or_repo_id, filename, cache_dir, force_download, resume_download, proxies, token, revision, local_files_only, subfolder, repo_type, user_agent, _raise_exceptions_for_missing_entries, _raise_exceptions_for_connection_errors, _commit_hash, **deprecated_kwargs)\u001B[0m\n\u001B[0;32m 354\u001B[0m \u001B[38;5;28;01mif\u001B[39;00m \u001B[38;5;129;01mnot\u001B[39;00m os\u001B[38;5;241m.\u001B[39mpath\u001B[38;5;241m.\u001B[39misfile(resolved_file):\n\u001B[0;32m 355\u001B[0m \u001B[38;5;28;01mif\u001B[39;00m _raise_exceptions_for_missing_entries:\n\u001B[1;32m--> 356\u001B[0m \u001B[38;5;28;01mraise\u001B[39;00m \u001B[38;5;167;01mEnvironmentError\u001B[39;00m(\n\u001B[0;32m 357\u001B[0m \u001B[38;5;124mf\u001B[39m\u001B[38;5;124m\"\u001B[39m\u001B[38;5;132;01m{\u001B[39;00mpath_or_repo_id\u001B[38;5;132;01m}\u001B[39;00m\u001B[38;5;124m does not appear to have a file named \u001B[39m\u001B[38;5;132;01m{\u001B[39;00mfull_filename\u001B[38;5;132;01m}\u001B[39;00m\u001B[38;5;124m. Checkout \u001B[39m\u001B[38;5;124m\"\u001B[39m\n\u001B[0;32m 358\u001B[0m \u001B[38;5;124mf\u001B[39m\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124m'\u001B[39m\u001B[38;5;124mhttps://huggingface.co/\u001B[39m\u001B[38;5;132;01m{\u001B[39;00mpath_or_repo_id\u001B[38;5;132;01m}\u001B[39;00m\u001B[38;5;124m/\u001B[39m\u001B[38;5;132;01m{\u001B[39;00mrevision\u001B[38;5;132;01m}\u001B[39;00m\u001B[38;5;124m'\u001B[39m\u001B[38;5;124m for available files.\u001B[39m\u001B[38;5;124m\"\u001B[39m\n\u001B[0;32m 359\u001B[0m )\n\u001B[0;32m 360\u001B[0m \u001B[38;5;28;01melse\u001B[39;00m:\n\u001B[0;32m 361\u001B[0m \u001B[38;5;28;01mreturn\u001B[39;00m \u001B[38;5;28;01mNone\u001B[39;00m\n",
- "\u001B[1;31mOSError\u001B[0m: ../../models/CamemBERT-base does not appear to have a file named config.json. Checkout 'https://huggingface.co/../../models/CamemBERT-base/None' for available files."
- ]
- }
- ],
- "execution_count": 5
- },
- {
- "cell_type": "code",
- "id": "8c96680645f077fb",
- "metadata": {
- "ExecuteTime": {
- "end_time": "2024-06-27T14:46:08.307140Z",
- "start_time": "2024-06-27T14:46:08.301542Z"
- }
- },
- "source": [
- "def get_token_role_in_span(token_start: int, token_end: int, span_start: int, span_end: int):\n",
- " \"\"\"\n",
- " Check if the token is inside a span.\n",
- " Args:\n",
- " - token_start, token_end: Start and end offset of the token\n",
- " - span_start, span_end: Start and end of the span\n",
- " Returns:\n",
- " - \"B\" if beginning\n",
- " - \"I\" if inner\n",
- " - \"O\" if outer\n",
- " - \"N\" if not valid token (like <SEP>, <CLS>, <UNK>)\n",
- " \"\"\"\n",
- " if token_end <= token_start:\n",
- " return \"N\"\n",
- " if token_start < span_start or token_end > span_end:\n",
- " return \"O\"\n",
- " if token_start > span_start:\n",
- " return \"I\"\n",
- " else:\n",
- " return \"B\"\n",
- "\n",
- "MAX_LENGTH = 256\n",
- "\n",
- "def tokenize_and_adjust_labels(sample):\n",
- " \"\"\"\n",
- " Args:\n",
- " - sample (dict): {\"id\": \"...\", \"text\": \"...\", \"tags\": [{\"start\": ..., \"end\": ..., \"tag\": ...}, ...]\n",
- " Returns:\n",
- " - The tokenized version of `sample` and the labels of each token.\n",
- " \"\"\"\n",
- " # Tokenize the text, keep the start and end positions of tokens with `return_offsets_mapping` option\n",
- " # Use max_length and truncation to ajust the text length\n",
- " tokenized = tokenizer(sample[\"text\"],\n",
- " return_offsets_mapping=True,\n",
- " padding=\"max_length\",\n",
- " max_length=MAX_LENGTH,\n",
- " truncation=True)\n",
- "\n",
- " # We are doing a multilabel classification task at each token, we create a list of size len(label2id)=13 \n",
- " # for the 13 labels\n",
- " labels = [[0 for _ in label2id.keys()] for _ in range(MAX_LENGTH)]\n",
- "\n",
- " # Scan all the tokens and spans, assign 1 to the corresponding label if the token lies at the beginning\n",
- " # or inside the spans\n",
- " for (token_start, token_end), token_labels in zip(tokenized[\"offset_mapping\"], labels):\n",
- " #print(token_start, token_end)\n",
- " for span in sample[\"tags\"]:\n",
- " role = get_token_role_in_span(token_start, token_end, span[\"start\"], span[\"end\"])\n",
- " if role == \"B\":\n",
- " token_labels[label2id[f\"B-{span['tag']}\"]] = 1\n",
- " elif role == \"I\":\n",
- " token_labels[label2id[f\"I-{span['tag']}\"]] = 1\n",
- "\n",
- " return {**tokenized, \"labels\": labels}"
- ],
- "outputs": [],
- "execution_count": 6
- },
- {
- "cell_type": "code",
- "id": "53310845f13e9d70",
- "metadata": {
- "ExecuteTime": {
- "end_time": "2024-06-27T14:46:13.910535Z",
- "start_time": "2024-06-27T14:46:13.689422Z"
- }
- },
- "source": [
- "tokenized_train_ds = train_ds.map(tokenize_and_adjust_labels, remove_columns=train_ds.column_names)\n",
- "tokenized_val_ds = val_ds.map(tokenize_and_adjust_labels, remove_columns=val_ds.column_names)"
- ],
- "outputs": [
- {
- "data": {
- "text/plain": [
- "Map: 0%| | 0/206 [00:00<?, ? examples/s]"
- ],
- "application/vnd.jupyter.widget-view+json": {
- "version_major": 2,
- "version_minor": 0,
- "model_id": "fc1766f4288040c0abe62a22f76de5b3"
- }
- },
- "metadata": {},
- "output_type": "display_data"
- },
- {
- "ename": "NameError",
- "evalue": "name 'tokenizer' is not defined",
- "output_type": "error",
- "traceback": [
- "\u001B[1;31m---------------------------------------------------------------------------\u001B[0m",
- "\u001B[1;31mNameError\u001B[0m Traceback (most recent call last)",
- "Cell \u001B[1;32mIn[7], line 1\u001B[0m\n\u001B[1;32m----> 1\u001B[0m tokenized_train_ds \u001B[38;5;241m=\u001B[39m \u001B[43mtrain_ds\u001B[49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43mmap\u001B[49m\u001B[43m(\u001B[49m\u001B[43mtokenize_and_adjust_labels\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43mremove_columns\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43mtrain_ds\u001B[49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43mcolumn_names\u001B[49m\u001B[43m)\u001B[49m\n\u001B[0;32m 2\u001B[0m tokenized_val_ds \u001B[38;5;241m=\u001B[39m val_ds\u001B[38;5;241m.\u001B[39mmap(tokenize_and_adjust_labels, remove_columns\u001B[38;5;241m=\u001B[39mval_ds\u001B[38;5;241m.\u001B[39mcolumn_names)\n",
- "File \u001B[1;32m~\\Desktop\\Sync\\Berger-Levrault\\Codes\\Ala\\Legal Concepts Extraction\\venv\\Lib\\site-packages\\datasets\\arrow_dataset.py:592\u001B[0m, in \u001B[0;36mtransmit_tasks.<locals>.wrapper\u001B[1;34m(*args, **kwargs)\u001B[0m\n\u001B[0;32m 590\u001B[0m \u001B[38;5;28mself\u001B[39m: \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mDataset\u001B[39m\u001B[38;5;124m\"\u001B[39m \u001B[38;5;241m=\u001B[39m kwargs\u001B[38;5;241m.\u001B[39mpop(\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mself\u001B[39m\u001B[38;5;124m\"\u001B[39m)\n\u001B[0;32m 591\u001B[0m \u001B[38;5;66;03m# apply actual function\u001B[39;00m\n\u001B[1;32m--> 592\u001B[0m out: Union[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mDataset\u001B[39m\u001B[38;5;124m\"\u001B[39m, \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mDatasetDict\u001B[39m\u001B[38;5;124m\"\u001B[39m] \u001B[38;5;241m=\u001B[39m \u001B[43mfunc\u001B[49m\u001B[43m(\u001B[49m\u001B[38;5;28;43mself\u001B[39;49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[43margs\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[43mkwargs\u001B[49m\u001B[43m)\u001B[49m\n\u001B[0;32m 593\u001B[0m datasets: List[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mDataset\u001B[39m\u001B[38;5;124m\"\u001B[39m] \u001B[38;5;241m=\u001B[39m \u001B[38;5;28mlist\u001B[39m(out\u001B[38;5;241m.\u001B[39mvalues()) \u001B[38;5;28;01mif\u001B[39;00m \u001B[38;5;28misinstance\u001B[39m(out, \u001B[38;5;28mdict\u001B[39m) \u001B[38;5;28;01melse\u001B[39;00m [out]\n\u001B[0;32m 594\u001B[0m \u001B[38;5;28;01mfor\u001B[39;00m dataset \u001B[38;5;129;01min\u001B[39;00m datasets:\n\u001B[0;32m 595\u001B[0m \u001B[38;5;66;03m# Remove task templates if a column mapping of the template is no longer valid\u001B[39;00m\n",
- "File \u001B[1;32m~\\Desktop\\Sync\\Berger-Levrault\\Codes\\Ala\\Legal Concepts Extraction\\venv\\Lib\\site-packages\\datasets\\arrow_dataset.py:557\u001B[0m, in \u001B[0;36mtransmit_format.<locals>.wrapper\u001B[1;34m(*args, **kwargs)\u001B[0m\n\u001B[0;32m 550\u001B[0m self_format \u001B[38;5;241m=\u001B[39m {\n\u001B[0;32m 551\u001B[0m \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mtype\u001B[39m\u001B[38;5;124m\"\u001B[39m: \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39m_format_type,\n\u001B[0;32m 552\u001B[0m \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mformat_kwargs\u001B[39m\u001B[38;5;124m\"\u001B[39m: \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39m_format_kwargs,\n\u001B[0;32m 553\u001B[0m \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mcolumns\u001B[39m\u001B[38;5;124m\"\u001B[39m: \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39m_format_columns,\n\u001B[0;32m 554\u001B[0m \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124moutput_all_columns\u001B[39m\u001B[38;5;124m\"\u001B[39m: \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39m_output_all_columns,\n\u001B[0;32m 555\u001B[0m }\n\u001B[0;32m 556\u001B[0m \u001B[38;5;66;03m# apply actual function\u001B[39;00m\n\u001B[1;32m--> 557\u001B[0m out: Union[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mDataset\u001B[39m\u001B[38;5;124m\"\u001B[39m, \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mDatasetDict\u001B[39m\u001B[38;5;124m\"\u001B[39m] \u001B[38;5;241m=\u001B[39m \u001B[43mfunc\u001B[49m\u001B[43m(\u001B[49m\u001B[38;5;28;43mself\u001B[39;49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[43margs\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[43mkwargs\u001B[49m\u001B[43m)\u001B[49m\n\u001B[0;32m 558\u001B[0m datasets: List[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mDataset\u001B[39m\u001B[38;5;124m\"\u001B[39m] \u001B[38;5;241m=\u001B[39m \u001B[38;5;28mlist\u001B[39m(out\u001B[38;5;241m.\u001B[39mvalues()) \u001B[38;5;28;01mif\u001B[39;00m \u001B[38;5;28misinstance\u001B[39m(out, \u001B[38;5;28mdict\u001B[39m) \u001B[38;5;28;01melse\u001B[39;00m [out]\n\u001B[0;32m 559\u001B[0m \u001B[38;5;66;03m# re-apply format to the output\u001B[39;00m\n",
- "File \u001B[1;32m~\\Desktop\\Sync\\Berger-Levrault\\Codes\\Ala\\Legal Concepts Extraction\\venv\\Lib\\site-packages\\datasets\\arrow_dataset.py:3093\u001B[0m, in \u001B[0;36mDataset.map\u001B[1;34m(self, function, with_indices, with_rank, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, load_from_cache_file, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, num_proc, suffix_template, new_fingerprint, desc)\u001B[0m\n\u001B[0;32m 3087\u001B[0m \u001B[38;5;28;01mif\u001B[39;00m transformed_dataset \u001B[38;5;129;01mis\u001B[39;00m \u001B[38;5;28;01mNone\u001B[39;00m:\n\u001B[0;32m 3088\u001B[0m \u001B[38;5;28;01mwith\u001B[39;00m hf_tqdm(\n\u001B[0;32m 3089\u001B[0m unit\u001B[38;5;241m=\u001B[39m\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124m examples\u001B[39m\u001B[38;5;124m\"\u001B[39m,\n\u001B[0;32m 3090\u001B[0m total\u001B[38;5;241m=\u001B[39mpbar_total,\n\u001B[0;32m 3091\u001B[0m desc\u001B[38;5;241m=\u001B[39mdesc \u001B[38;5;129;01mor\u001B[39;00m \u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mMap\u001B[39m\u001B[38;5;124m\"\u001B[39m,\n\u001B[0;32m 3092\u001B[0m ) \u001B[38;5;28;01mas\u001B[39;00m pbar:\n\u001B[1;32m-> 3093\u001B[0m \u001B[43m \u001B[49m\u001B[38;5;28;43;01mfor\u001B[39;49;00m\u001B[43m \u001B[49m\u001B[43mrank\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43mdone\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43mcontent\u001B[49m\u001B[43m \u001B[49m\u001B[38;5;129;43;01min\u001B[39;49;00m\u001B[43m \u001B[49m\u001B[43mDataset\u001B[49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43m_map_single\u001B[49m\u001B[43m(\u001B[49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[43mdataset_kwargs\u001B[49m\u001B[43m)\u001B[49m\u001B[43m:\u001B[49m\n\u001B[0;32m 3094\u001B[0m \u001B[43m \u001B[49m\u001B[38;5;28;43;01mif\u001B[39;49;00m\u001B[43m \u001B[49m\u001B[43mdone\u001B[49m\u001B[43m:\u001B[49m\n\u001B[0;32m 3095\u001B[0m \u001B[43m \u001B[49m\u001B[43mshards_done\u001B[49m\u001B[43m \u001B[49m\u001B[38;5;241;43m+\u001B[39;49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43m \u001B[49m\u001B[38;5;241;43m1\u001B[39;49m\n",
- "File \u001B[1;32m~\\Desktop\\Sync\\Berger-Levrault\\Codes\\Ala\\Legal Concepts Extraction\\venv\\Lib\\site-packages\\datasets\\arrow_dataset.py:3446\u001B[0m, in \u001B[0;36mDataset._map_single\u001B[1;34m(shard, function, with_indices, with_rank, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, new_fingerprint, rank, offset)\u001B[0m\n\u001B[0;32m 3444\u001B[0m _time \u001B[38;5;241m=\u001B[39m time\u001B[38;5;241m.\u001B[39mtime()\n\u001B[0;32m 3445\u001B[0m \u001B[38;5;28;01mfor\u001B[39;00m i, example \u001B[38;5;129;01min\u001B[39;00m shard_iterable:\n\u001B[1;32m-> 3446\u001B[0m example \u001B[38;5;241m=\u001B[39m \u001B[43mapply_function_on_filtered_inputs\u001B[49m\u001B[43m(\u001B[49m\u001B[43mexample\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43mi\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43moffset\u001B[49m\u001B[38;5;241;43m=\u001B[39;49m\u001B[43moffset\u001B[49m\u001B[43m)\u001B[49m\n\u001B[0;32m 3447\u001B[0m \u001B[38;5;28;01mif\u001B[39;00m update_data:\n\u001B[0;32m 3448\u001B[0m \u001B[38;5;28;01mif\u001B[39;00m i \u001B[38;5;241m==\u001B[39m \u001B[38;5;241m0\u001B[39m:\n",
- "File \u001B[1;32m~\\Desktop\\Sync\\Berger-Levrault\\Codes\\Ala\\Legal Concepts Extraction\\venv\\Lib\\site-packages\\datasets\\arrow_dataset.py:3349\u001B[0m, in \u001B[0;36mDataset._map_single.<locals>.apply_function_on_filtered_inputs\u001B[1;34m(pa_inputs, indices, check_same_num_examples, offset)\u001B[0m\n\u001B[0;32m 3347\u001B[0m \u001B[38;5;28;01mif\u001B[39;00m with_rank:\n\u001B[0;32m 3348\u001B[0m additional_args \u001B[38;5;241m+\u001B[39m\u001B[38;5;241m=\u001B[39m (rank,)\n\u001B[1;32m-> 3349\u001B[0m processed_inputs \u001B[38;5;241m=\u001B[39m \u001B[43mfunction\u001B[49m\u001B[43m(\u001B[49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[43mfn_args\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[43madditional_args\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[38;5;241;43m*\u001B[39;49m\u001B[43mfn_kwargs\u001B[49m\u001B[43m)\u001B[49m\n\u001B[0;32m 3350\u001B[0m \u001B[38;5;28;01mif\u001B[39;00m \u001B[38;5;28misinstance\u001B[39m(processed_inputs, LazyDict):\n\u001B[0;32m 3351\u001B[0m processed_inputs \u001B[38;5;241m=\u001B[39m {\n\u001B[0;32m 3352\u001B[0m k: v \u001B[38;5;28;01mfor\u001B[39;00m k, v \u001B[38;5;129;01min\u001B[39;00m processed_inputs\u001B[38;5;241m.\u001B[39mdata\u001B[38;5;241m.\u001B[39mitems() \u001B[38;5;28;01mif\u001B[39;00m k \u001B[38;5;129;01mnot\u001B[39;00m \u001B[38;5;129;01min\u001B[39;00m processed_inputs\u001B[38;5;241m.\u001B[39mkeys_to_format\n\u001B[0;32m 3353\u001B[0m }\n",
- "Cell \u001B[1;32mIn[6], line 33\u001B[0m, in \u001B[0;36mtokenize_and_adjust_labels\u001B[1;34m(sample)\u001B[0m\n\u001B[0;32m 25\u001B[0m \u001B[38;5;250m\u001B[39m\u001B[38;5;124;03m\"\"\"\u001B[39;00m\n\u001B[0;32m 26\u001B[0m \u001B[38;5;124;03mArgs:\u001B[39;00m\n\u001B[0;32m 27\u001B[0m \u001B[38;5;124;03m - sample (dict): {\"id\": \"...\", \"text\": \"...\", \"tags\": [{\"start\": ..., \"end\": ..., \"tag\": ...}, ...]\u001B[39;00m\n\u001B[0;32m 28\u001B[0m \u001B[38;5;124;03mReturns:\u001B[39;00m\n\u001B[0;32m 29\u001B[0m \u001B[38;5;124;03m - The tokenized version of `sample` and the labels of each token.\u001B[39;00m\n\u001B[0;32m 30\u001B[0m \u001B[38;5;124;03m\"\"\"\u001B[39;00m\n\u001B[0;32m 31\u001B[0m \u001B[38;5;66;03m# Tokenize the text, keep the start and end positions of tokens with `return_offsets_mapping` option\u001B[39;00m\n\u001B[0;32m 32\u001B[0m \u001B[38;5;66;03m# Use max_length and truncation to ajust the text length\u001B[39;00m\n\u001B[1;32m---> 33\u001B[0m tokenized \u001B[38;5;241m=\u001B[39m \u001B[43mtokenizer\u001B[49m(sample[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mtext\u001B[39m\u001B[38;5;124m\"\u001B[39m],\n\u001B[0;32m 34\u001B[0m return_offsets_mapping\u001B[38;5;241m=\u001B[39m\u001B[38;5;28;01mTrue\u001B[39;00m,\n\u001B[0;32m 35\u001B[0m padding\u001B[38;5;241m=\u001B[39m\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mmax_length\u001B[39m\u001B[38;5;124m\"\u001B[39m,\n\u001B[0;32m 36\u001B[0m max_length\u001B[38;5;241m=\u001B[39mMAX_LENGTH,\n\u001B[0;32m 37\u001B[0m truncation\u001B[38;5;241m=\u001B[39m\u001B[38;5;28;01mTrue\u001B[39;00m)\n\u001B[0;32m 39\u001B[0m \u001B[38;5;66;03m# We are doing a multilabel classification task at each token, we create a list of size len(label2id)=13 \u001B[39;00m\n\u001B[0;32m 40\u001B[0m \u001B[38;5;66;03m# for the 13 labels\u001B[39;00m\n\u001B[0;32m 41\u001B[0m labels \u001B[38;5;241m=\u001B[39m [[\u001B[38;5;241m0\u001B[39m \u001B[38;5;28;01mfor\u001B[39;00m _ \u001B[38;5;129;01min\u001B[39;00m label2id\u001B[38;5;241m.\u001B[39mkeys()] \u001B[38;5;28;01mfor\u001B[39;00m _ \u001B[38;5;129;01min\u001B[39;00m \u001B[38;5;28mrange\u001B[39m(MAX_LENGTH)]\n",
- "\u001B[1;31mNameError\u001B[0m: name 'tokenizer' is not defined"
- ]
- }
- ],
- "execution_count": 7
- },
- {
- "cell_type": "code",
- "execution_count": 8,
- "id": "a654180a-a536-43fa-8984-c7b87b419f93",
- "metadata": {},
- "source": [
- "sample = tokenized_train_ds[0]\n",
- "print(\"--------Token---------|--------Offset----------|--------Labels----------\")\n",
- "for token_id, token_labels, offset in zip(sample[\"input_ids\"], sample[\"labels\"], sample[\"offset_mapping\"]):\n",
- " # Decode the token_id into text\n",
- " token_text = tokenizer.decode(token_id)\n",
- " \n",
- " # Retrieve all the indices corresponding to the \"1\" at each token, decode them to label name\n",
- " labels = [id2label[label_index] for label_index, value in enumerate(token_labels) if value==1]\n",
- " \n",
- " # Decode those indices into label name\n",
- " print(f\" {token_text:20} | {offset} | {labels}\")\n",
- " \n",
- " # Finish when we meet the end of sentence.\n",
- " if token_text == \"</s>\": \n",
- " break"
- ],
- "outputs": []
- },
- {
- "cell_type": "code",
- "execution_count": 9,
- "id": "6990d89800dbb440",
- "metadata": {},
- "source": [
- "from transformers import DataCollatorWithPadding\n",
- "data_collator = DataCollatorWithPadding(tokenizer, padding=True)"
- ],
- "outputs": []
- },
- {
- "cell_type": "markdown",
- "id": "668dcf9750404d1c",
- "metadata": {},
- "source": [
- "# Adapt the model"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 10,
- "id": "7bd0cddab7ddb448",
- "metadata": {},
- "source": [
- "import numpy as np\n",
- "from sklearn.metrics import multilabel_confusion_matrix\n",
- "\n",
- "n_labels = len(id2label)\n",
- "\n",
- "def divide(a: int, b: int):\n",
- " return a / b if b > 0 else 0\n",
- "\n",
- "def compute_metrics(p):\n",
- " \"\"\"\n",
- " Customize the `compute_metrics` of `transformers`\n",
- " Args:\n",
- " - p (tuple): 2 numpy arrays: predictions and true_labels\n",
- " Returns:\n",
- " - metrics (dict): f1 score on \n",
- " \"\"\"\n",
- " # (1)\n",
- " predictions, true_labels = p\n",
- "\n",
- " # (2)\n",
- " predicted_labels = np.where(predictions > 0, np.ones(predictions.shape), np.zeros(predictions.shape))\n",
- " metrics = {}\n",
- "\n",
- " # (3)\n",
- " cm = multilabel_confusion_matrix(true_labels.reshape(-1, n_labels), predicted_labels.reshape(-1, n_labels))\n",
- "\n",
- " # (4) \n",
- " for label_idx, matrix in enumerate(cm):\n",
- " if label_idx == 0:\n",
- " continue # We don't care about the label \"O\"\n",
- " tp, fp, fn = matrix[1, 1], matrix[0, 1], matrix[1, 0]\n",
- " precision = divide(tp, tp + fp)\n",
- " recall = divide(tp, tp + fn)\n",
- " f1 = divide(2 * precision * recall, precision + recall)\n",
- " metrics[f\"f1_{id2label[label_idx]}\"] = f1\n",
- "\n",
- " # (5)\n",
- " macro_f1 = sum(list(metrics.values())) / (n_labels - 1)\n",
- " metrics[\"macro_f1\"] = macro_f1\n",
- "\n",
- " return metrics"
- ],
- "outputs": []
- },
- {
- "cell_type": "code",
- "execution_count": 11,
- "id": "ea5d16f59728e2b9",
- "metadata": {},
- "source": [
- "from transformers import AutoModelForTokenClassification, TrainingArguments, Trainer\n",
- "from transformers import RobertaPreTrainedModel, RobertaModel\n",
- "from transformers.utils import (\n",
- " add_code_sample_docstrings,\n",
- " add_start_docstrings,\n",
- " add_start_docstrings_to_model_forward,\n",
- " logging,\n",
- " replace_return_docstrings,\n",
- ")\n",
- "from transformers.models.roberta.modeling_roberta import (\n",
- " ROBERTA_INPUTS_DOCSTRING,\n",
- " ROBERTA_START_DOCSTRING,\n",
- " RobertaEmbeddings,\n",
- ")\n",
- "from typing import Optional, Union, Tuple\n",
- "from transformers.modeling_outputs import TokenClassifierOutput\n",
- "import torch\n",
- "from torch import nn\n",
- "\n",
- "class RobertaForSpanCategorization(RobertaPreTrainedModel):\n",
- " _keys_to_ignore_on_load_unexpected = [r\"pooler\"]\n",
- " _keys_to_ignore_on_load_missing = [r\"position_ids\"]\n",
- "\n",
- " def __init__(self, config):\n",
- " super().__init__(config)\n",
- " self.num_labels = config.num_labels\n",
- " self.roberta = RobertaModel(config, add_pooling_layer=False)\n",
- " classifier_dropout = (\n",
- " config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob\n",
- " )\n",
- " self.dropout = nn.Dropout(classifier_dropout)\n",
- " self.classifier = nn.Linear(config.hidden_size, config.num_labels)\n",
- " # Initialize weights and apply final processing\n",
- " self.post_init()\n",
- "\n",
- " @add_start_docstrings_to_model_forward(ROBERTA_INPUTS_DOCSTRING.format(\"batch_size, sequence_length\"))\n",
- " def forward(\n",
- " self,\n",
- " input_ids: Optional[torch.LongTensor] = None,\n",
- " attention_mask: Optional[torch.FloatTensor] = None,\n",
- " token_type_ids: Optional[torch.LongTensor] = None,\n",
- " position_ids: Optional[torch.LongTensor] = None,\n",
- " head_mask: Optional[torch.FloatTensor] = None,\n",
- " inputs_embeds: Optional[torch.FloatTensor] = None,\n",
- " labels: Optional[torch.LongTensor] = None,\n",
- " output_attentions: Optional[bool] = None,\n",
- " output_hidden_states: Optional[bool] = None,\n",
- " return_dict: Optional[bool] = None,\n",
- " ) -> Union[Tuple[torch.Tensor], TokenClassifierOutput]:\n",
- " r\"\"\"\n",
- " labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):\n",
- " Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`.\n",
- " \"\"\"\n",
- " return_dict = return_dict if return_dict is not None else self.config.use_return_dict\n",
- " outputs = self.roberta(\n",
- " input_ids,\n",
- " attention_mask=attention_mask,\n",
- " token_type_ids=token_type_ids,\n",
- " position_ids=position_ids,\n",
- " head_mask=head_mask,\n",
- " inputs_embeds=inputs_embeds,\n",
- " output_attentions=output_attentions,\n",
- " output_hidden_states=output_hidden_states,\n",
- " return_dict=return_dict,\n",
- " )\n",
- " sequence_output = outputs[0]\n",
- " sequence_output = self.dropout(sequence_output)\n",
- " logits = self.classifier(sequence_output)\n",
- "\n",
- " loss = None\n",
- " if labels is not None:\n",
- " loss_fct = nn.BCEWithLogitsLoss()\n",
- " loss = loss_fct(logits, labels.float())\n",
- " if not return_dict:\n",
- " output = (logits,) + outputs[2:]\n",
- " return ((loss,) + output) if loss is not None else output\n",
- " return TokenClassifierOutput(\n",
- " loss=loss,\n",
- " logits=logits,\n",
- " hidden_states=outputs.hidden_states,\n",
- " attentions=outputs.attentions,\n",
- " )"
- ],
- "outputs": []
- },
- {
- "cell_type": "markdown",
- "id": "77f4fc68394aa754",
- "metadata": {},
- "source": [
- "# Fine-tuning"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 12,
- "id": "79161ed938cad895",
- "metadata": {},
- "source": [
- "training_args = TrainingArguments(\n",
- " output_dir=\"./models/fine_tune_bert_output_span_cat\",\n",
- " evaluation_strategy=\"epoch\",\n",
- " learning_rate=2.5e-4,\n",
- " per_device_train_batch_size=16,\n",
- " per_device_eval_batch_size=16,\n",
- " num_train_epochs=100,\n",
- " weight_decay=0.01,\n",
- " logging_steps = 100,\n",
- " save_strategy='epoch',\n",
- " save_total_limit=2,\n",
- " load_best_model_at_end=True,\n",
- " metric_for_best_model='macro_f1',\n",
- " log_level='critical',\n",
- " seed=12345\n",
- ")"
- ],
- "outputs": []
- },
- {
- "cell_type": "code",
- "execution_count": 13,
- "id": "931792b554582a9f",
- "metadata": {},
- "source": [
- "def model_init():\n",
- " # For reproducibility\n",
- " return RobertaForSpanCategorization.from_pretrained(modelId, id2label=id2label, label2id=label2id)\n",
- "\n",
- "trainer = Trainer(\n",
- " model_init=model_init,\n",
- " args=training_args,\n",
- " train_dataset=tokenized_train_ds,\n",
- " eval_dataset=tokenized_val_ds,\n",
- " data_collator=data_collator,\n",
- " tokenizer=tokenizer,\n",
- " compute_metrics=compute_metrics\n",
- ")\n",
- "trainer.train()"
- ],
- "outputs": []
- },
- {
- "cell_type": "code",
- "execution_count": 14,
- "id": "89c997e6a944bc70",
- "metadata": {},
- "source": [
- "trainer.model.save_pretrained(\"../../models/Fine-tuned_CamemBERT-base\")"
- ],
- "outputs": []
- },
- {
- "cell_type": "markdown",
- "id": "05f2cf51-2a76-4de6-9b22-af7dda1eb805",
- "metadata": {},
- "source": [
- "# Inference"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 15,
- "id": "e48a67b7-814d-46d5-a079-fafc5b6adf86",
- "metadata": {},
- "source": [
- "model = RobertaForSpanCategorization.from_pretrained(\"../../models/Fine-tuned_CamemBERT-base\")\n",
- "tokenizer = AutoTokenizer.from_pretrained(modelId)"
- ],
- "outputs": []
- },
- {
- "cell_type": "code",
- "execution_count": 16,
- "id": "4cb75af5-234f-4863-94f8-45a364e1b877",
- "metadata": {},
- "source": [
- "def get_offsets_and_predicted_tags(example: str, model, tokenizer, threshold=0):\n",
- " \"\"\"\n",
- " Get prediction of model on example, using tokenizer\n",
- " Args:\n",
- " - example (str): The input text\n",
- " - model: The span categorizer\n",
- " - tokenizer: The tokenizer\n",
- " - threshold: The threshold to decide whether the token should belong to the label. Default to 0, which corresponds to probability 0.5.\n",
- " Returns:\n",
- " - List of (token, tags, offset) for each token.\n",
- " \"\"\"\n",
- " # Tokenize the sentence to retrieve the tokens and offset mappings\n",
- " raw_encoded_example = tokenizer(example, return_offsets_mapping=True)\n",
- " encoded_example = tokenizer(example, return_tensors=\"pt\")\n",
- " \n",
- " # Call the model. The output LxK-tensor where L is the number of tokens, K is the number of classes\n",
- " out = model(**encoded_example)[\"logits\"][0]\n",
- " \n",
- " # We assign to each token the classes whose logit is positive\n",
- " predicted_tags = [[i for i, l in enumerate(logit) if l > threshold] for logit in out]\n",
- " \n",
- " return [{\"token\": token, \"tags\": tag, \"offset\": offset} for (token, tag, offset) \n",
- " in zip(tokenizer.batch_decode(raw_encoded_example[\"input_ids\"]), \n",
- " predicted_tags, \n",
- " raw_encoded_example[\"offset_mapping\"])]"
- ],
- "outputs": []
- },
- {
- "cell_type": "code",
- "execution_count": 17,
- "id": "314d7a28-ef8b-4367-8e6d-79a55951fcec",
- "metadata": {},
- "source": [
- "example = \"afin de vérifier le kilométrage , pour les véhicules équipés d ' un compteur kilométrique , les informations communiquées lors du précédent contrôle technique sont mises à la disposition des organismes de contrôle technique dès qu ' elles sont disponibles par voie électronique .\"\n",
- "for item in get_offsets_and_predicted_tags(example, model, tokenizer):\n",
- " print(f\"\"\"{item[\"token\"]:15} - {item[\"tags\"]}\"\"\")"
- ],
- "outputs": []
- },
- {
- "cell_type": "code",
- "execution_count": 18,
- "id": "b0164373-ee70-4d63-a0c2-ac2197b37b22",
- "metadata": {},
- "source": [
- "def get_tagged_groups(example: str, model, tokenizer):\n",
- " \"\"\"\n",
- " Get prediction of model on example, using tokenizer\n",
- " Returns:\n",
- " - List of spans under offset format {\"start\": ..., \"end\": ..., \"tag\": ...}, sorted by start, end then tag.\n",
- " \"\"\"\n",
- " offsets_and_tags = get_offsets_and_predicted_tags(example, model, tokenizer)\n",
- " predicted_offsets = {l: [] for l in tag2id}\n",
- " last_token_tags = []\n",
- " for item in offsets_and_tags:\n",
- " (start, end), tags = item[\"offset\"], item[\"tags\"]\n",
- " \n",
- " for label_id in tags:\n",
- " label = id2label[label_id]\n",
- " tag = label[2:] # \"I-PER\" => \"PER\"\n",
- " if label.startswith(\"B-\"):\n",
- " predicted_offsets[tag].append({\"start\": start, \"end\": end})\n",
- " elif label.startswith(\"I-\"):\n",
- " # If \"B-\" and \"I-\" both appear in the same tag, ignore as we already processed it\n",
- " if label2id[f\"B-{tag}\"] in tags:\n",
- " continue\n",
- " \n",
- " if label_id not in last_token_tags and label2id[f\"B-{tag}\"] not in last_token_tags:\n",
- " predicted_offsets[tag].append({\"start\": start, \"end\": end})\n",
- " else:\n",
- " predicted_offsets[tag][-1][\"end\"] = end\n",
- " \n",
- " last_token_tags = tags\n",
- " \n",
- " flatten_predicted_offsets = [{**v, \"tag\": k, \"text\": example[v[\"start\"]:v[\"end\"]]} \n",
- " for k, v_list in predicted_offsets.items() for v in v_list if v[\"end\"] - v[\"start\"] >= 3]\n",
- " flatten_predicted_offsets = sorted(flatten_predicted_offsets, \n",
- " key = lambda row: (row[\"start\"], row[\"end\"], row[\"tag\"]))\n",
- " return flatten_predicted_offsets\n",
- "\n",
- "print(example)\n",
- "get_tagged_groups(example, model, tokenizer)"
- ],
- "outputs": []
- },
- {
- "cell_type": "code",
- "execution_count": null,
- "id": "54213be5-8221-40a2-ac60-04670d83babc",
- "metadata": {},
- "source": [],
- "outputs": []
- }
- ],
- "metadata": {
- "kernelspec": {
- "display_name": "Python 3 (ipykernel)",
- "language": "python",
- "name": "python3"
- },
- "language_info": {
- "codemirror_mode": {
- "name": "ipython",
- "version": 3
- },
- "file_extension": ".py",
- "mimetype": "text/x-python",
- "name": "python",
- "nbconvert_exporter": "python",
- "pygments_lexer": "ipython3",
- "version": "3.11.0"
- }
- },
- "nbformat": 4,
- "nbformat_minor": 5
-}
diff --git a/results/BERT/legal_bert.png b/results/BERT/legal_bert.png
new file mode 100644
index 0000000000000000000000000000000000000000..d4b4efe0501748329313152190ee4ae114d41077
GIT binary patch
literal 32101
zcmeAS@N?(olHy`uVBq!ia0y~yV1B{C!1#rOje&t-od&xu0|SF)iEBhjaDG}zd16s2
zgKuI<K~8>2PG*uqS!z*nW`3Tro`s%~jzUIBNkOrdzJ4xTfnI)5y1v5ie@7S?7&r?&
zB8wRq_!B{xafSWzOa_LUOP(%{Ar*7p-Yu^PDZO><V|CgjPkj|d-boCMOUhq%tSfZ9
zD5|aHwR!EL)oHp5R(JGn)j7R-QG{4wY@o|V?Jf?Dt|pV0vP?@RwtTTGzGv^`)}|uh
zG->99_s@4dQ!0LTX6ADrKc{y;O(!Qxv@w7{gZIqXSP+AOL16(ii19#-;{}Lj*ultT
z1!m+mH1L5K4bm_rY@<pT7#L8(r9pbOL1NRbEt$c~pWFTW@z`jl&)4V)vv2ON-|wR~
z`Ptdo&tC;)U0LQU?K{KZpz8X&BKPJWD^s%%W?(QeVX|Vl(s(ju)z#depPsJM6P+|o
zKfdnL`^&%I?|=VF==_O4KR=shU(*Sn{xIcG3#ZXcA1)rAkifvesmIIT-g;W`KfdB&
z>t(yT#KeVi^?wTMj=j2gYN~eW$D`t_Lsy4=`t~g>Cufb`?l(qf=iA@k$p7A&!9<5Y
zT%3Wy_oDUtJ&#NN@BjaI{WP7(pw#N;=jNV&QF^c9G4IW{->huE%m4q;zRYK4k=y$E
zZ9i|{|F^98yzTZ=-yECSb}pFp%eekuP42xtk>BrCufP0l#|{foQPGo^FJD$It*oe6
zv0+2N{Q7@8*X{ilwfy+K+V8QI)zzza?AYOVTu(>m%9odyRXv~2udiD>zg*2eIDK;~
zw|Lf>8HTT}uAcsG=X1HO85f;?e|vkG-~P`6^LrJ}=64F3m+adAo}uBE;sRy|**g`F
zd;PxG|9mR08@Xx8sc*ufVq&MRU*CRheX-iXQ-=>b%gOzjWq)r&;^C67SHpMj>Tydp
zn)&4I+qavpfB*F9(ga24FZZh7hjPC7{QUg&6@iPtTy&Q=_0nZ!W&QN|bMeh~(RUA0
z_SehR|0yhedC7JEpHJOu!d72Se);lcZ=f85LrEcf16%RGpU+oEZ_kU`UAFd|^?RG2
zzka=X*e*X!rS{j$<x-|uQx+{={PgM5r9XaD*#3MndDG1tzvC8#dvc!6cqATI;kbVP
zzg>?WKL)8NK5rYIeQix)RbFZG1z)wv?{>Y`tNnCR{nfR#v-j-Tb7gCGxNh8@9bSJ=
zPuCCMU-$RP+qY|f{HWOb?N;_yx$FNP_uIdkmAx)7Cuhx`J$poig%|&Rw|l);_;XHS
zwG;R6$G7v#Uwd(J@zeM3*H6=pe)aG7`}LpVQqt4254CV|ad8C|6l^%wFTeiHn>WGN
zimY~(zFx+yw`0MnvfX#rg{_UUZ2b*#ZZkW7SVqQ*nwpv`+w<dVe>`koUei14-r>s6
z&rYToW&Z#7w=Z|=PQP>Z|7&W0KAmn=@StJw>58YO!Ivezyx;#n?r<CL<oWaGH#Ikx
z{(L$;`}VfntJii`f6w!sVc>Z1;6W}fu1iNcg+n7EDo%Y~Q*xNs{Kf0_`^$dCt&QC5
zw!7?Y(Uw#5G_0+oL5VFXDe1}6r%S)Rz5V)T`ux%@HSh25kJpXfHbupEzTMrzd;RwR
zb}X@9x7xP)+lxuwdVvlO4afRqUst?2V)&wLckKV~_5YW@ySw}J?c341(c7+^n5g{q
zcK&|LZ;_i)I@|bUS3P);@bmZY*K>-`75=fA?^9X1b6(9S&*SH_W}D@H`EZzj^^zq=
zHlANPJ#JOm+gqW(|2Q_Yt&HDa=Xd+x_x=A@=HK7<>Ep*h508#@yWd&K+EiS~oIm;R
zzm@0a*;bdl+xa|<^Z6tZIXS(TSKjYd$-TcXHaxy|tCN!xC{pYG{(8Cjyj`(X+|#Z9
z4)fc;_;x!#JR@TT6Em}E-W>~B+o~&>_wU#J_EmQ435bZ80g8=nxzWC}OjaKI%+Ro`
zZ<#1VhQrL6Gk5)X)LnYF^!m$V()l4SE-g#C>;wM({X2EW42}Bw`r7}0zgNB4cznw2
z*{h$=um3mY_+0z?eV;ykD*63(`{@*;mph-&yF4{KZsp$U@AEdb7Cb#QHEK)7#CLl>
z`?0dHyf`Wze<d_LcIlEODy?m8UX!k^k6%Ag*<DLZ%j@!T|LJGbE(`nHJpA$aPl%Xa
z%!)&YobDDLm*wK+z543v>aUl)^<Q3C=p419VBzI`X0t&)xzNDK{N)h0{)^r3_g&uq
z@2mdN<Huk3S-+d0BH6a^wEq4W?eKM1;{X2&FMW3>QojDr#$)~R@vEz}o<4o5sjR&C
z)VB*40%rQC^-3DMfucSrI9OCt^5vI#s_N>mcfDR$ob!AC|KIk1pX%3Nys<GkG%9M<
z_xttn*JFx%_x*Up-797KYRjutp1JELyUSI&SXx?ImAshn@9+D1?@53D{>`>|R(`*B
zdAI()h_|=4UVhwfzi!bYrM+J+c}p3kbX;2-9j>}3uKMlPtlQhxrk$Cwu=Mq{muJoI
zhg4MTm{<GlX4Krnpn%-}@9TO{5GqHUoo^pM*QzvVzFqCAH}Me>D;^$hfBoU%;az{f
z-4>OWUOm75Uu5udzfdo)uC%kWmVz?u)$sVMvgLOcYOmYj6dqT(RAr@mzg*wrDbuHC
zAL$U3vMN~-SMjiQ^0%0sKcCGu&A(?8Tl4YguBX$YL*nAr$=Cl01eIUEzP?_r8@=tt
zC2#%EemUDOM}+-PoIdTnyZrq%fBU~Br#|n0IxYIkx7+!zU#(t$?f(CN<$r&@UO&q)
zxounS?PdM;{~{)tq@9uIl{WVS6;)DAg%1ue{{3`X-*3P0yt-eRw!dC1-u3NPcIj!|
z?JrJDRIYltbb86X%IBGXe|_cR<12d;efj<TS4WN>&3t)jDJX=aw&%sZwOd|we|yf&
zMT*XCD?(R?%`!^u`t#?{mxJu`8DCyp{PkjSf5}nNa0}V@>vq2bl~bS+`R$F3&g=L6
zTE!=0@nD9rlC5oQ{lCxiuWm~9_MK_udhp=ETl?$xi|Iy{ocb=yFiVG-q2QRB|Gb<z
zKeS7Ke0bQVl6-ECW#~739i26s-hR8CZ(dVd+xzHI((!E{A0L-4EG+E%{rlTnV-?AT
z3m3YDg-uKTy?lOMmWPqu|3AgYw`E>l7P``NQsUOA-n+ZYt+(ag*>Q0~%APNmy!&#e
z$Ck}}`1*CWj9tx*jk(v?#kRG#yH8CHTfXb_Y{TTXM5CEK%<Oy?bLXo(?o^*waBF+{
z`+IKp_tnaNxBvM>*m=^&JH_WMw@sZsUHWhvud#|`{@$<GJYPp{%bBUT@>sN&S68A$
z+s5SMeOm8Z6D5w#Ex(uf<M(gtxz{_NM?^(smAt-o_Q&P<|DJ5I)zdpCVPCiBab4+s
z@xul^+wWDKo{;k9_V)G1t{e2cT0c2(v72SA`c$p;(;wfTAGfzk^5MgWZfR-Dl5ZdD
zmFC@iGp9$|e4W?a$!fkuR?kGkV-j21+QKT=3iry}-}~_SvoX(MgLV6UWj#K`t#86}
z*dWJjwxFbBrq6u4yMgyVpSOQsac{D}UFMIUKP9cp-<4S1o1z)a(|t5aMO9U@Em6X~
zPsUJ1^71m@)!(+NN(L|US@~__<NK@M@Be>qLdu`F`Sq(et-qUR&U4rRlnJFL^(edd
z&H1#ae|zF#Hs?u&r**em<nI6f?>DdKvG4cm^Cj96pFDXY5f>MC43uh@|D9L;&T{$u
znSXzOmlo?jTJh`U^10t`Z^@ke@ZrOUTmGi(uV3|f{{KJ6KmLB$#I0u_SNkRK@p1Wj
zlN__zm+RE4@9XdR;Iz2<`#Znu%kJixv+>L2l)SibaO2ufQSWbWUmqA1^{Qmwq#l3!
zzgG-+4*&S^xS!Ya*euiRtWrLP0<#6()(j>J`uh65b1XKVDwCI&KYo4c+Fgg+c!LuY
z7jDbHUw7(ppQN#yZq$~7JMv~ZGgLC)-`l&qCeO*K>CxlIvfrz}yjXbPfWz_PT(jBx
ze!sKsm9veyc=2N2-&@=B*LUmh`!VJC^f~9(FIeCpDJdysQ?Y^P@P%JrU$1`i|N8!a
zS6}|yka&2`xu>hw@0--#-oDs1c9M=*_q%<+-$h*we!20u+~rBC-mC67IXS6#%37C&
ztdHBfCE*~`@!}k_*>-<EILDUVO!a$S`sccPzg(!FUtb%q^s>^|*XD-#suwT2ezG$t
zC}>OKVKz`Bw(wlgQPYBpuHr9Ouiv-o)0|iP%xBJ=dFt%hwO?P~D>%#>wXbGpP;l_q
zzu)g`o0@Jt_U<jWxZafwiHEN&a^+4<NwLV|<mB9RGw0XCcKNKoyN<`}oBn*UxIe{c
zrb+ItEn5Ade6A~UrvCeK+223AY3Hf$#)gI$Gk=3(*z?l#_&QBbE*_pMN5$jU@Z0@x
zXiK#C7QG>%k&RDg#f=*go|hK6cBgE=UHW$Gb*sJR6DLl5bzx!iG~MWDU#v_)Rq+zf
z$y+KuCe5w?_cLl|(Na+L&MR%^BWqQ1;k^C-np4lyQ&V65e!u_v*6VSvcZ3-m8BLiv
zbETlNTgLTuu~s!dHe4>w+x^yTxtx9Zox*k-8ylJT(NBN>J1QQ(V$GU0Vb2*Ed}SGB
z8KeSMhpk;DdwFN-X|dSCqoSb3RzkvpsBJkfXC&yUt3RJ|eC?m+y+NrF_Jx1aa_2?=
z{B!F0vgONHuUO&n_V)Jmo8n7vubMn%$`g*psgF;!^ULe4eDyQ5@8{Z;aw`|D&JPI*
znY!-w^Et(SYhSIrvbVZiRYfIar|f@QDdV)Bx?eBV-DgDXtJ!(!@@3PsGZM>_Yd;<p
zm$E2WuxIaHQ6Zs6FaAZ(ejlGZ_ho6(B4&0zk5&I_U*Fle*~rK!Bp~3ymcKshwt7u{
zd_Y_$VnO-++VJP|s`GlUUs@Nt8x(X~b8cRGIc>rOg_Ylgr&U%|T)7eYyQ2Ero4}Hi
zEtlEF)Im;Y<B?nxoww6fP*8B!yIrqArTQGBHS!h(4r!Zrp4#c==B8y~apU9P6*axh
z?EG0*SB3WNy0$L%b;;egx3+q>C9ZI8=L@|zKlk>w(8~O*Uw?xq3o5(4n4P~bu&iub
z-H(UtzVmD%+4*EHES(-#b?STV_jh-Vc%1v~eq~I(rDgxe;-iGDRY}3F_wV1YU$H{t
zxDC(YcK++3;jv3!US6)PqvJC@zHaBes@J;9KmKMY*p|p+#bC1FakPVWYI=J1j}H$)
z2~61EX5o(?6_wT1*>`u9R=rr*o^{nJePhI`yLC0Mqn&NGo8_3z*4y<$DK$0q<-&Hk
zpyhsZpFDrQ{BS!zsC6BBZvC>VnA9^f3};!Dt^&1&s^9Mom#_KI*woZyQuM^*xSjpK
zAB+Een!f+Vp3mnlYp>t4XvGSR%BreYYq#GkiZb41W$L>r{O*#9^s2d6+JEhc+p}j+
z$<wLfuikFIf9>Ah>bYv7))VjMU3V3aUBWJ3v*Ffp>#{cyudkcTe!X1w(DCE$_J2Nr
z68<v%{eObG#r3Z>v2uS|ZvWTwe%<f4n(H6l@>x@HGj;mO>({}qV&j(s=clyIk)Y;L
zvi$D5dXrBsS-=0^u3Oh_J~FR)Rrmi&+xqqEK?OtN;WkYj9iPj~{jXn-udlsiDX+Eu
z`0w}o<CWcdR+P=ZRd*o#WQx(t!~FJF^!NW+bl(1d%>A0ry=uO*mel`!9sld;^msq3
z>r4zLp)4f~87ZK?O<LMAG2N&opU+vJ-&lKpU#;GnSNsnjKK%6U+qHLhcdxbyySq&|
z_tq9q>+*M3Zs+e`Tln~x<u?7uO-s(1hWaH;oj?Em7x$livevI^&dl19dpqp!@9)=}
zSh-8C$ChVaT;wWcTeW4+o;@<ZYrb4`2N`O9zvl6lziYSOvwHebJ9@$9&70?{eX-hS
z8fvxo^%672fK?UpOJCjJ{`U5EP<pAWtJ`XI?d7hAZPF&??_y@2O<R8Qo1cih{Q5tC
zYTUakeY2|1Ef<oPU!TA4r(0xXq<XHa`Rud1J{;mMeb_3V_3X^ds%JCPznoC+&$ziM
z)%2Q_xVZZ6yX#_im#r**ey;CrW@hH8bLYxx>{U*lKkq+1u4*MH#mqEL&pOz|DrJ^4
z<FdizA3rK2+ZJxiy?yOhbC+?bWpj;}SJ#qR_rITtd|_*A>;H4_>MhyV^_GVUfBO7+
z>ZD1Fy!Cb-$=u(_%%1h&Kx5Ufm&;A-|JC%$T8F*g|9@ZJ&ZpBZ+sQK&l!JP-Zece+
zmDX^lWM;06t9&XNKK03wq>UPiijMPZK6Or-JX!X3`TKioL5U&k{Pa2J&l~Xkz5oAj
z{M_taPbPVT+M40<wO2QuP22o(>GZf&pAG%f&de|bMKdV0Qd3ewUf#>w7?EN$b57LW
z@9*!g_qY4`WXkb%GXLxUM7(NhV%q!n+wIUe=T&!0A3c8_zFO~nU|=9<K;T~8@7(37
z=gso(t;yg2cbk{NDu3~qf<{oOz4y<j(|*QJ=D)a=z5eQ|(A8f+P47w_YwPHzPoGYD
zGGpprpXs-Dzgo3AMWSt)|NOY88&7?=)7D;nDtT|!*GWqzmu+8HW1oI@*3&Ju&(6%8
z6q35_zhUw*4NJ?&{Cz){ndRMyc)Dx%r&HRg8zW3=e|^!imEbuXkd~I#yLHZ~OG~|1
z-;J}B+fx+b;`_e3qGAWA^9$;f_ZighQFiYOx%uta>-E;Xy7gbLhKI(*y_@m+(Tj_V
zL(gqC1C7F@Zdt(Wpe$`uu_0<p#=?(}kH5~Z|2^ABZSuX!=X0xGC&kHwTD!Hs-);x>
zG~Vra%y%=#4CG<Zuz*?4je<+h%U)ga{P^+XuK)jjm;U&W$jZj{YH`1v)mp>T>bu@-
zI&G42LIBhz{{8)Z_;kJ4Eg2UVU6yPA`?%kJ)s7uIPJNTJt+LR)2&yGPg96w8e!std
z!h{K0x=oASdM}-sX>3~b#N%$kVO}qP|LbpVZoay&w)*eC-|yAeK0IIlZ*x*|a_Q%@
z=C2#s<t7|Wx_D=2F{ll$s;X-F&3f+ZCHeRFMQuoE6cQFLeYf-Zsr&cir^l8>Ze9QX
z&*$^6A9d?rTj1DS^7*W}^|VNCE-tIg>z|EN&1Qe?umAIS%5U2b2bh1o*?hj_<<jX{
zhgvva-QK>Qi;HVZ=4Ca|&;zK2pX~caFLqbP!6w#MH#RPIa&o$|Hrl-O^|iNKUhR6&
z#9eaP*Zd`@r@Ah7H>jk&`L^uukH`IBsq3-jFFV!eT>y<P1uk~e)YJ2O`SNAe)2ZQC
zc9mxH%35tXmi+zA&EUt!`>(%VzyF=YrVUkJvp~rcRDuM*-uUHs#9vTJC)Vv69$Px~
z^5c&mKUTe3x!k1mRmj`>`{Nh8_lG$+G<<t|dpW4*wEb?`?05VB|GRAU-wRaf+}f7A
z`n29|pN}6uYH4U(0QE0HrQ`E?)$1NtS>60?SN$!ASJo=z?(XvIYq#G64Krx!>iVt@
zTYG7tbNiOU$8O87>*(rUZ55AOk-dIzn048k3!tiFp>z8y{raEXDlb9R14snaW&HT@
z<Ehi9FH7g|S$N>U0k*#eoD4CmCZ_kYH*5~_J-w#%?X9i1{?`9`sh<4cKx12ff4`ut
zY_83Z2hGRK@7E-I7=hYNH`Xspz19~$ZPg^dh+iv~rCw9>pBM9Vr@9_%nnc^il9xe?
z%irH?yP9SC=<(xT^ZPZIfBe4xKd+*?`uYdGKZ&oet#xy9YI^woeSS-8tK{~3RjZRN
zw?Cf#$Er?q-Q&>k>a?_Fg_nKJB^w(X-2wt8fZ4CFuU|hs_1davqpTAX6tC({41H<;
z|7ZWjh0g4UPoF;h;>ya&AFoB{henr|zJGgbYj!UqGn+)%>Z={Sy{2jB=0v_V%HF!y
z@6+ecz0!F*4s!7EnW<Q=oB3*5q%qIoAMbX*?@@Gald!Y1b4y5AP<S(Sx@7eBywY3C
zKgIaPM6KCnwL19n_iq)ZvF9r*E00}S8QlBr?QQ9>)mJ&Vxs5@cf;}G&ar>Pv`upee
zc~BpYclFg*Uutb`X3zg&oGKye{mu6OpU-Z~{pRwlzM8cq=jNpyKOS}WZq2?f)fV`1
zbHUH2)1@z7ya)=hMT-~rzFxn7->IO2tD4%{#>@QY=WWToeeJ`SFD6MzNj)orms{z&
zmhIV?d|YZ$PfJUSRR7P5?(&k(?R=(x{`}#wx3^y$yxi~DlarHsWv$CBzFFRUKmXO9
z@As;Ym0picKXUxIw4I$@+xG3-FD~^Ke>~A$&akMsc<#5gFKf+t4&T_E-fx(9$3of2
z=n}}rsI6J4F;y>@9&4AcdvRm=C#k57Nv(yqbGJ+CMr|?p^Y?G>#^mExy8pN5-L-oB
z>-Bo`q~v5?eSLjp8=E^jzFhL|y<hix?X~Bu48FZ54eqQs$;R;Dnp?lzTHWYvptkcm
zJyr){&=7_19E(8EI6_Y$J3~kUj}^m$C|6fkzwhO5Z!KNF@0Zu}`Sts<xtJIjEKVD;
z$ub12Eqi+_^H2+C*7bFB`D82>gvZy0hOORu?CE+228Mz(ju#9KS-morLGSl|pBEGy
z3>sB+Y-W2oGksoQJD+S!tP}&ofyhon*xbf3=xhgKD&yD!bK|nzu_sTS<l^9%a5AN+
z)Dkr75p%?V=jqd@npRd(X~wzR-!(McJENkmuC1na>}o1VL4hzkzg)=cYina;*<2j*
z%b1v$R6_p<u(7f2+93r}ux`@q*{f&FkZ^ZvcrAEg_RLdxZyOsMSxP{1Jhi3DJSHF(
z!vS7_3ydIE0W(C7;SNLcg+SfeXP0f-WE4KV@p8)6s7NKyBu&O5ju!<<5^X`j!BgkX
zefugXfIV$<WbW;4Q?F)~mTH0w@e^j0_0u|BSy8d0rnYwG#f_XBZ@=~O^9%Fx>PkvZ
z{&^2%l`jvYET7rzvx4WGot<N2`CctZJv|MisIqdWjje5LENI@R;qrlmP>Ya0%kN*m
zmKLh6Sbb{UTeI0`k4^<yV-d?z65@FM#j@`AvJ4Ci7fJ*%{J<#dd#{C4IK^n@m+Slg
zZ9SDdS>1oy$&^bioWfH3H$)lC^a;&#Zs#kt0d@7VKRh`2>Fd|ff`SdNxmZCaKHzv^
zp?&1Ykr#iz-wzKAoR}P4Sh(@o`T6T_+_=G3`}WeMOU70IqVxAIoiat__*?MU={M8s
zF^jF=?Re~SnuURZ;lVSI5no<jb~iLMjM|bhG5Pi)*KQGM>DBl4R$IT_UGXsqRMvvV
zpm}+(f?6)N)!!`M7QeWl7+e4MD;Fo{#QF2*gND!I>;HyM`tsv(|LXAdaaEtsnqS?P
z8(mpl?LBGIq)8<|pH4r0|9*U9V<Tu7Z0fXWWv6a6H#dWZB6qa5v{=Z^wyWLsZufh;
z()aiFo_o*B@Zd?^jY=2!KV<=ByC1#^VqiEc!zinsD$&N}eOmf&n#A(|?w6K$PX2N&
zI^Xi$`MK8N!s>o&HpM$Sg6dSzh&5;=&1m-7U5~o7Q?}n$_5Atsr<7gI4nBFiHRtVq
z`$(JTU1{dGdr=WHsqFo|wf%O#Ji5hnFSYSXgSy5)pHA1-*7nv8TXSKOs`r-i_i??F
z#>@ELTQSVJE0*tf@x$}!FQ-1;{rb!AefuQZ7G4i~b8*Mc)n6a9?|=P|-+a}XH2bf@
zj0_nHa_jUM-mP76W|>}p?d2EicYo~^{Iqr6y{DCrd(E?6U0Dg52vB*se16@f-|zRY
ze|mbl_1p5Rq2ZQy=UbPri`tg6a&7eXRngn?X4%*8>yxouWSV`=B5L#P?fLJ&%(-+a
z=<oOa|3gnt(^d8Ko~CngneXf^`S<OP9y#JM>EG}B|5s&SUl+AGt@oVm_d8Or_x=0z
zI(1`&%=iD__y5oQ^5Ww1(sv9OLUQcpEUvzHf5D`p-SeXNIPaG~bCLV{w<-I>AOG`A
z+Adx7>U8d`-<Rz?C!JK$jVh1W>6%*gv(4_)w=HuXaX38I*nD&A%YweTyR(;G{cTjb
zWrI)9QPb$1J64@cndCWX{q*)lOnXHopR8S>knhU+TSIHbqq>Ia?tAr)XGyemoys|-
ztrs3wIJ2nxv~2uy(dnxfEQy&Xwz<Ye_tUnhCyy)Nl)cHW*XL)gs6KK3w(~r(TRe?k
zEG2VHK>-uE*e&zko=PJg=f!TlK@kxVQ~#E|y|whXT(!^sf4^2685zmwKKAtVgf<5k
zfRb~mseNEf%pB!8-j}xJ-WCxSe$4TF(v_v&;;iiKpg9+H-WUJ>ecwNI!h{Dq<V8hG
zqn`6J_`0Umb>`Kluk||h_LYq2in}i>qxPtVR;N9Gy><4lRk7ax-BvDMa%b6Z@qg>i
zo2z+GlH0jt^R1SjnV<fY{>qEqb?j{HTJz)IgBNU1UwnmGX8oD8>+jo(zMQ`2^LLlN
z_x`u%gA!N1?dgA=zV(mb4d%W5tlztSH`>iRrK=b0mpF41xA)Z8kY#TU89fyVy~E{o
zDl{N!-65gbTwbR`3*OxS^@`!~+sv?r?aP_=*or)OtEtbw*+y3_@=RLgEZ<Etd_QfO
zk!m<w<jLceD-!Zu*?(&M+QF~D@zWw&;6h=L-8`|}+N2imxb1qNUgz$2ySyivRDXMu
z^?m1?O{Yb~#g~KT7B61Bxasy=P$(vEGMAU1AH1(`{`~m4uU7k|q@+w0t75RYJ>~YB
zUD^MtWY6gZ%?^m)TmRbOt=**c*7svq$G-fz>(A5qCa-tJ?b=Z6(tK@Vc+9f>|MJ(?
z-I}ZE-|cxm{%-K%>g|(M{+0*T-~IJ1KWeYO_x1h%y6?@q5})`ljyH;5YTn(vjBKxU
ze6qSz+HdOccc1*Y{`toB%{4Zn;;*7#i=`bl*kZoeVfUp^qW3TPzm58JI%!?>>k`kM
zzS8{9UR+OZta;nGSWByz@5#Y^^Q11AxiH^uT~*2*@6J0Z#p8I>riFJN6)k-xqOH2z
zL(}r8*ICW5qOyra-KRqXvM#^-QuEgD{QBUzriQbhh^%AYV=MCFt)jg?|J2)BQ8&B2
za<#NJuiSBF*`Z5a(T<)wAKP@Fexs$QeR1NRNmf@Aj65GQ#&Pj~2-~oJc~k9Lu5TYh
zK&%?AciTU#-4p+@bVC1fn?CNZZHXG27lgF;iM_mV(dqd9m6obnTH5wyd}X5Y<#wu1
z&3;xyi9C4w@`Yf)Pq`n}l_ydr<&-&>Eq-LRMY8QpnRD6UfRYXUaxynp#_eOb*bRzV
zwaG8H?4Gsz(us-6p;1v<lTR)YR9;oLVZzLrD`l<AUi6yZyRaegu!^U4__~r?%ia6s
zK*jacsZ&cYg?)Jpnnq&hm%DPXnO%16i|Flnp=D*;c9py=YD;Hj@V##wpL>1P-kF~^
zM}7IVqmJv-*O2==#D4vGw{X^<cUEg})`yty|2zAg_lp&~|6hrF_V|!rS>noVD$lb%
zKa|P!KRbE-)kS6t>$<qVhQ_RzpX6NgG}-nm`|g*OM`!)GCY9@XZtDE2i!9B*UfsY~
z|D{dphF-|~yJi88pRc<%?MY0mar)GIJ9q9^7C0DLmbmg;n#c<8kc74EEHWo_F1Tm&
zzr4LRQ)}zZr-}cRb)wIxRm%O|6`Zm_@5Pm%zqi{?oPPYpe#?1Hoy+{Y`!as^DMxMm
z++p;|U-pmFpEa+fuRHyH6tyCLwm@mN=!-A6*f(>1{XCCn?XAfx=YLPp`n$v6W%kw;
z(IpeAw(tdgzvlSrqE$qp;y$C<zMF3LCa*mwy8VyP>`mO>TkSNX_rE)R<=2v&sZ*@a
z*IqbuJ(&4cTb9G_u7El}mxjydkG#~<&<ywT+jyy{yPx&DJyVUB-^QZu(|xRw{;a<v
zM4!BV$+4sH-n>uWzfG8b)p=i7_n$plhqlkZs&qg3|Fx~JYi6EZ_vca3)9F*RzeimE
zvFB0I)M>Xv^R}<PlXfD_JS1&<_@8X!k}V5DPFlKpom*$oKIKi))LU=7PU{9ny?e>}
zhasPz^>@{&%Q<01eiOS+`_=PxJu=#zeB<ED7c!zA4(?lL`F&!_I+(OE<My`PzV-f~
z0qNGZwpU-T$A^c6Oqn!kl1b_*k>$1P_W!H0{d~q4)aL`u?5$X#u{M5x+{KF*rA#sc
zwyvK&VZwrUcXwyMxv_DUY4)^Jr%#tYpIe^w<HN&O*Vo6-J+H2*=@}Ur>E-1$>1^8N
zHeP9!(4Nlk-}C3JTfi`@&1iOParnF7)nay1S8Z3t9iMgWNkZ5{Gud+&=UcVTH(8(?
zRer+o+VA-5-$IUeFdR>Q$at^n`?SWYSEtwhdcTVK*Y|SktuG6<Rc+(D_UwsFuIIU{
z3m3bZ%HC2mx)5Esa9<bKr!a2&)lom9t{ytSN+5o9)Q`$jIYuuV`ORM?giUn%yJnX8
z#{Ijt^Q|hHlee~M%ZK&NC70{Nvg6$v*VS!RT2gMY?wRq{mjQjtbbiI^)~tOcA-JGe
z>F-rVB?nn$CaGGTe=n^jL@rQ`zPu$RdsdxwL5SII#eDzGY{%PQ`p-S|b-CZeF21sN
zozvoj)~z>N`r`1`_AOQMVQb@_KewF;s&0Qq=}O7SFXOh_dgbTymep4|k8iupn<_q4
z(|Es&`zP-$cAC+rw=K_(y%Z>GuepEz)kVch%qhlNRmD?o_jEC2n1|#Um#FM_iDX^c
zS3Fr|b$jIAD0XeD8`0U{<gY%=dH(vstHiJx?|Ru2Isu6*eOG@^&;0t&{@nTNSrZdi
zcx6|V9{RIZ>VjFw-`i~&4!Z?PdF(ek)G68P^Ka$<9{#)F`>)#vFDI@0QTcWAwVx7Y
z)vl+@z5FH`ee#|X8?kJ0Yw9V>b*t-+?q7ObPCGQNaHoo;oxq9|qtHKknRgGluZ{k(
zDs)5tqNcr1-#>k5Qg`q7jq73EKW|w4vA2skp{*BQ=Q;CJzwGstRX;wQOfox}V{|fU
z+8MRUPJdt5owL}z^ZS1$2U!+2wrf9T+|0QKYBufv`!*j`L7lUFF0=Q?qi(DJe?H6p
zj#wK88Xfxe`~LrR-g-M-o<4p0>e5p0-q>eLyW;oFulu#~-M-&>bIb2l8qGc{Wm&W&
zuIlAdW2@<)76fP@?Qz}HNrk7U=@ze&WSAGdCpQ02gkY1s)$<8yF*ltjI^N^`6!lZ}
zY1Pl)cW3P`n(vic*7_^bG_gSXu8D@<!F}iIOY*FyEMD1d<~ixzzu@ZLH({mocvrVa
z`p4Y!y&KK?a-zT4>)p98K~03qc7+S~y^@k$ac*Lxmge^C`m{#A6tS%Ow8l?er>D+}
zUUTKnqu^_D(vMdE_j|S8EYXxdH0%1Mc=>yV9*_TRj@o_cbLrXa)n5+2ZJuKnu|j+I
z-3!X|b1qGL661O4)Llux{_R}f0*`alZ{__Q9<yS7l5_1<wJZK>n`%?OTI@C4rux(@
z_1SbS<^3)*e=Y?zI|^s|9!)KCFMqtqXtr11Rw=e$AzNm#S(fsC3-Jq#Jo-H)dsf-p
z3$|7#mOuLVe*5y*zu$`9-|RP$@mKKjg3^4^CGWQ_ef|2ZW}o2|waF%HoV{bOE}yEz
zytjXApVX$C+RmF79txU$?ak>+n`B>lo!7ote*4mwyW6@h+%Y@^s`5{r(GN{?bX~D7
zvvd#N!+*OL1n)Ta{zlN>TP<_G{MFXoo?TqjS9g2<Qq$l4S#KK`pVHDO<+>x6Ho593
zOLjob3bv!|tNe4gE32bYHcQ_sR*By)f9m2g=bOL&793^wR_~MkZZP}F^U#Z5Z{9!c
z{M>~Nl)QN}q*wizR@eJ!;jNn|a?DnyHcXFVyHaQ2we#${JCBNFZN+YRpMKlL_bn{s
z!pr}27{8uP*-=z0Y_jXlofuYDR!w#F<z5qX_4KaotNlIa+Bdh$hrSDB%-eTs$|CFg
zvFe_iqRwb-y(#+O?ZV@FZx^lpy=~og`#*2XoaNVAFL}Q2=0gc>sTXe--bvPV-51uq
z?!El#+3njxOE&Zyee(V?eR`Od<?;yeGOp{=kAKbG6k723V9wdNZMzqpE?BKl+RGO-
z{e=VT&k*~>*@+UT^mH!&JHw=!D*DPJ&Nah2DkAD*zT?bIT;5Y-BUb5Mbo_PQ)HeHX
z`=5(VLB2D#yGiq>X<MC$?tUjzbWFagW}jhX`Tp&3bz$Ah;x0x0Hsac1tm|(b^m)!M
zpZO<pCM{~N6%l`>9lyFg`OL<|kUmGV6pvH-df{n~D~&#RO$iT*d;Ia$slz#!8p;&C
zUdSwLIVYHP_mI2wrB7Q*qN}#*UY~la_2<h~`_`7G&JXM2vicgMA9U0**Z1VrPqOZQ
zRXHmkH8%y+ufEm#^Yx$Ns8@fAqvl<_&~nP~S@w7N&?3KyRma0hYQL{e<!?8d9qSp{
zx+&^RU*5wwxilw>JLgVrVvK2SxOYb2!uh-!{>Ecd=S8oP&8(CXk`G&dH_-q1JmsIi
zPJ3&E8g*}<Z@qlAHv08H{`%|3nr}{8^7(w`O3CoD^`#oCPTpdUob~T#*xx!0_eodO
zu7vl;-Euu_z3s~M*+C}hN8@t6S7wNQGnidcxjxM6l-7k=f8DR@o$;T4$8?G3xvqJ!
z5vx|m{f(^d6Il_Q(qQ-F^#ZH;?w&h+IgP_#B!o>A(tovLf9#Egr?TJroW9<-#YXmo
z#)ZZ;PO7P5uN=Z0=N>!eC3?R*pnmnCKYK-9ynXp%(d(y~>s~B-ojGy)^2j#D<2`q$
zZa)>dM>~42uvPo(KYb6mKCS)vr(bMJ*3omyzv`-YZBu!^qW*S3?A`5Ke;r|)Uubbw
zPv`3SYNODoyzmwKLl@S?aeZF@LhdZjJ%_sj7w+;i?wzT)VE)}7a!kGH<yG_6RsR)F
zeKP&ko<7~LhfHln<Sy*6FK1PKDrNgy{jJye>kZo!tG-WT{Ip?4Xx`@6XQ%!)DL&~p
zb=qyw2TvEe>%U!;{cGzqm8ny1&s9q~rM>ROqsl#}bW$ZQrAtr#`m(J{_2i$uHea@L
z-_HK?z|2oLL$Yn5sCEB~t)kUiv+|cE@7_PX-EO8$#M{HK^LfgXt!IR;(2oBb>Hf*f
zWJ&Fcd$TU=_q&_$eBGVp>icf2n#LFx_3^;Zt@B#08t}-T)GlQC_NZvKkCElIgX?~5
zf7Cbss&nGMu-0GUE!P#~{iAXjuXHrLtK)camWT20n`7SvGGumN{@maC*?yNw`>)8a
ze@tz^EcfHx?GvGT*r4S6Z|N^LkEOJ-SG?@r&%5>$<H2`dZ0_#=P@ZdXZI;}Nw+j~;
z%|4a0{D1uZlPa;d&E3BDTRfi<d&FGk%(ZOoZ!fkpyMKKXdRO|bZKlU;{^uEIvR8)$
zWL<s{cX3u_?)uBGOeY41Ec-k^9#m7#Doibpy>GvbaWyMw?dEQIwN7S+9aFE&Nvhp#
z8an-3N7xC;GjDDOe3LtQoi{(E<@33~###TSud8}<`RuKEvzB{)uDjfFFxTYiBfqV2
zWvkZR`&YT&?)D5F{>`47kAGXa?d!hUiUzw~$s6wdDO~?=Ys&KEH9JBCcl=#jzvlMq
zr<q}?cWeyzwX;{ep2uw+V*R?ZwCX6!>#NeguYEjoCwP09W7XGV`(~Ns_uAc@u;zq<
zyzkB(-@H}MMs9R>=RSSK?oRymFS&OcrZO?iGGek?wIJB_^3m^n@1E>DdfT+VbW)y-
z!1CMe+k+2i9(z0O_M58etBhy=l(lnt<^HPpo2a{|c7lymLC(svc^UO-jb%%ot!!N%
z%6z+Z&KBzr`+AsS{~y!$I<FnL`~<_^>eCFZKO(=Le4<#D_mS`H-^}^X{_cri{VnCy
z(bv_l=bX6w`r6#kfK^-B`?Aub=2^M1WO)3(khR-!z3qgh%lEB`2r7D|I=3`$_5`W^
zOvBkFzduyO-qv@^tUSu~;af<<_Vzz(<@BV@?}yC2zA?$>KePP5b(Y%SG^}n!KW5u%
z`SkOo&D|#DsYiK!2fLMJUC5SVxUfa=!ovqIBR?K5PE1@_Q*$oINO6Z-X6DKzudX++
zOI_x-?td}$X|;-{zW(jCKUYuwvfX~`bzbI2t7LcY)_b|)-^Y(>xwd8}XQ-_8KXuLf
z{`+gS;-zb*&-&DLddtg#zHQ5&WL9Ub+qV2!W_r}T%TJ=8yRf`?mY7>wxjtk0$E!g{
zf1L;QkuMc_Uokv%U+|pj$=jC~9oKukC{gzCmjj1-Z%v%G{B&I0O7?#>eZP)uT@exX
zGD73q=S?RLCr$f%<C4#Q*%c9&GVeAnzBN1W!IxQK{F~Xfa+j{rmlb}ry62D3-JkEQ
zrrzf5eI>Zm)A`nwSAWiztW6Bru`IqPsJQFvdR1%Fx0`jlIKEu#ST|9<_S_j&>(HRQ
z@c-KscWo%HP!8o{$mo)2zISF$)*MeSudYp-H;c;2>BYy#t4+&vdQ-N0(&WicU%p&<
z<3_~3@`9On9vSui_|ETjQZvlyImg?xC$8U97CmUJcO{&^$HIN_F87luMw+`fgq+vA
zxA6OQm6HZq)4ZxOQq+G|MuPHbikT<(e=CpL$60xbcA|$3OzKM~&HGt2o69d^dc>qX
zwr5`n$Sv6G{z<lfJI{wppqx_Xu%@YYt=B1?E2Zn#2Nn5kyw(2u_1<XqYZiRJJ>oj$
zeiSU8_TC~W+4tVOw3)ukv|g)ti)Ndb_wZae)_l{dWXp+{8^1mL_sL-P7F$#QkGg-q
zyo_~z8h<)<en36X;R}c5^RHE}p8Vu#;<YQ?)|qb(?8r(Nx<75wYVWu|fo+PlMwJ~^
zM_J744l)EVA6u|no~g>Xc$K`<VS_2NX03Yj=FKsEzVtIWW~*~=Z`%}=soMP7W<$YZ
zv30gf%ogtZ^=e+?rl>PfA8tGTtv;!Dd`XsVa#N}2c8|UP4#%a6g?#z5W0ill@7^Vo
z5B<?@%74_v8@1)%hN$fyE?)o6W1TAg^OZ`v_{)X;b=UiM8>x7%ZK}OuY85bd5z|#m
zBZs+*m_(%?X?NdTGG~#|-u+gd`G2>>uiSZbl8U^4_qvzYqrA5-eO*vC@lo`zmwRus
z%`4hgcG%#{Ieqpo2lLsdMu)6xnpE_3>Xh)Ans<+1UKD3|5Y(lveH^vIKZpCcP*_`H
zMtRk~efQ_BTYCHQ)^7&_Z}#ajz0KbGV)O1clb16u&fZltJGTDs{ixdQy0Jk=+tsI@
zJ|~<1=gw77FWv2H$E7#YzjiDNVOSx^c=xZsh1oN2{S&yAV>Wf_RMVu?75rhVL(`0t
zr5{?|&-cH)q;mhslu33wYA0O2@uF!`(cX`b<3K^QeA;8ahySXwO3ZmP#9x)H+4IOK
z_S2(&s~Mq7w5zvYI%BvuE4DpRB5KFC3G@Ectp4P6>gm(OYt72mnU5ZrEte^ejmo<&
zzF+<ISNU&Q)mihBt9KjN@2$V>(8_+TwzcI_(QYoUQ&Z!E)&<RuEmYjMRjz+daKtkC
zGTDo-Y9n{;h+e|Boqu()><am(X1njj7Y7(me-m<TBmXQti{(q@B(J)@J^c6U<##;2
zwV%T)qkMkp74CTC^fxBPkikM3l28uwD>k%E3XWLT{co?+%SzQ<sTt9KA79eD`fmEX
z@HOJo-(A}D*1hEM0gJmfTXK7|z5Z{r_rAVwn^&20)yH4=etDmN5qxdpyy!J6yUj9R
ze(Rcc_TI*WRS##)>&kywAN4$0qOI=M+?D^=toh;a@c+8kYfmUl({~BAbWN?we5rT#
zcjf+5Z`*9nsDZjStG}madKaxz<H=im{Qsq^S5uiBL{pf*J3Ewbytz63_0Q+?!?Uth
zRegK2a(Y~q=ilGo*BAbO^Crjk(+OqJ6vL)Xn?R$Bk(<+APV&}UdG+;2XO0C6?`%(K
zwu`HecpLTW<?8*stGA~fOVhi#QhfTmpoZ=3V)v`(MX`q4cP>4f_u?{dyXoK0Yfouv
zO}?50Dm$;_KHoaI_2#LI9H5HQI6k+yq<Z=i27P|r`Uf?wzvkDxtzal%-PHIyF-=k&
zI`3|peols0!oXp3+SyATg34QxkMp&*htGYwb<36z@z|1yhYlb9`mkMoRp{z4&?tRV
zbF=jR3I>aF=Fvf4KmIywG<%8ox|fmbvm5@u^?&>I@jB(G`BT>Z7R$L%y`8Uw>yD<S
zYt)8k3+y6_15R3sCLiCumFvp>buUYAou0#(z9`naey`=fn7<o7cg*`$w7P8B)7SG_
z_*#k?9M)cJdgsK<#&cok^LeYM$Ja#$9o4yGXlS_W_q*Musi&rB{XRd_*!{fy|2^CC
z@4t%>dRy{QQ?lXLw9tT6_W!F}jXrsQz5ejytfIo@RrS5hGg)^%`Ck{}xY8F?oNk|g
z-A}tr?qc=o$zT3{E0LXT?W<d3Tl)FJwq3OqHCv6dn0uGrNeeG3+nD5B8ym~TaHS*X
zK^?~ny`ZQlt@wEGz6Q{EPx9{{A0JPhJ9n+W{ogI8iuHCr5o&E|IdT1ZIH+3<8l9KU
z+u_J>`z4_Fb=>qRQ&ucla-`7S+&tXS(D07{(+czQv+>LI>-<5z(%QQ37qnitU!UH(
zU-k;K{69VZ&7Qi<d!`hv*%~v6Z5#K+uYZ$G7iFjH-FA<ep}-837k2H~?cwIumSZ+s
zNl9tWHg^}7jNjkhUbRbqe{b)vf4|?Gmc5Cn{dzV0*52yvpjG)(r%uhhwI%baUHY*e
z$ys)_ReP`VGfevO^u6Y<N8d#vJ!@IBAQ{d`hM>HTX_`yO}fyEA3clsQSQKO##@
z?_ZND&O4Lo&%n;W@Su<5g>T2*yzq*O9dY%4OCxo}zP`O3-p((7?els2_+*FqR;8<i
zgoS0lYe!#TeBd+n<{Qw6)TAZW=ljF^uN~z5Ri!rB^Z%YzyB{6?3mUFg-tB7kf}M$h
zfuXmz;hj=qV&bL<9WQV1;#>2kO<Q*2#EGo=!lI(8{{H^U-@Tb@T^^Q|waUxeJ9TT+
zYMbMyPkYD3`S>v}_*c7I)mPK<DA(uhS)W+nN<TGa-eWtpBJj9dr7i<Q!?9yLR(eTW
zqpohxkJpXeb!Fq(yzSX9FD(U+)18+zO6e$je@}P!-FG*(6+icLb#)DuKF7E~@$B!)
z`6@5l7z=xTvoSENU}q^2_ww<%0$M1s{eE5a?QOZ2LCfx&nwUVd)ssx}?pXNFwYs|T
zKS*@V8lAhv=WTE8EMC6v&nIu&&u5Hh*;H;??Z@a~y!B^6-?caPTKt<m&G|Y?!86I}
zth;vXzA(cux#Y<SLECRPlD&L=LqY3rD?TP2KT_!F>1j0k{|_EjH8nTTDml<Z;nx=z
zo!3Tfy=43S&f<2tsuQ8h85Yc(_5E|5=cN^Lfn5Aw%5n`E7#Pwyr$uI`MR+kDn0e}i
z{>2B9_g7y@yMDZ!FYNsdv2}W!3=CHoZyxx2;ixP_LDaqYxl4j($J!=0aXx2YVEEe6
zz{jv+tI_P(_kW5UCYcl#NM9FYW?(3ornrEaAuIRevyi1vr>#hBn6CFy?&+%{1_p+6
zS}Y|D4r@dG_{_^3)~u2HBhJLY5HSD8Yc7TW?fvqnuH9mGf0}9dl5JWTGXukla<J-<
zZS&%9DMvkI)LF~Mzz}dBG+!TVH7R_NzS%O5Nood`G7Jn3%?b;c8T?-EP4-^>wU0OK
z{EUR~=^(d+DS^aSZMn1Vzif0=@3mqEhK4q02X=-zSFX<vo-!|d(FX?o*gL&TuQD(&
zNO3XBG6X0eHZUoFYUg>Wm-%r-6fXnA0>+Qi#TXX2y<cm$qPM;1$=l3pU;jpeWfVb^
z)>pmG>0L68&z-DdYp`f?6kEm49ZC!g4V$GwrnM$YRQ-7DdrDt3eEy9DwRUC(h71;v
zzeKyQ$zJ;;>z?w};PuYc)u)oocJ5LEX_I4=Wq6=syFKEv{e30RUFHj$Z?*s44L0&w
zV*?*UfdP;6{~sxvi@Hx=-5WX~Px{)qtE!9)3?|YnB@7LRr_T=h^7fiZ<qY4dcdz6=
zZTN7Xg@M5#`Jui(LxT!vg!HoA+84L%el~r3B_sOcQ5ORP!z^(oD+Y&2CZJkFN8_rc
zAmgkW$7*l@UU&mqTJWpPdljhTwAr)j-z>RJRUhs%gUo!aZ@`f8%<EKc)tAd>CoQ>s
zt{aq(*XeRFFf5p<w1C;cczfCZz@<;8X`H$5{avl30Gyi7iZRME%-Shmp6)T}+-m`~
zeE;t5)r<@b8T&Y1FkCpZ_D-nah4wOz;zCsh28OesK%BL4G3(M>bLJ>}O5K{*z{kMw
zz>B4XA>&xxt;vdqC!Fi#2&h=Fef4RO8P`B&^jzF4%fN7;R^S5Tf|*7<&h|nV7#SGu
zuz~Dd`FCck)tp$B)koJb<~qlK)+sn_c6VTJNXvS<PI=N2yTXH?wpQ%`ufG86U-jmm
z-5O_a@o-Q<)Y-tta5iY?*>$1SMyj695plc>3<q>U>LhPfXt-CZKIQuKG30xg<2pTd
z1_lOD7;FyNx#!k{9|4hP)WpL<OEAD@Tsn1ERQSloJyHw|4ZjsYK{Yu(Xx&%Oo0FC#
zzMZMS$G~tP7G&(qpQ--KPQ7hQa;_1Te)Srhiwbx^9+^3HxAw95m%%>W3sG7ae)PxZ
zM4k0)3=9v#zyY!3iAd)8iA*znFFmLL?a6JhRs`7<v~y3?_iG2aK7B1IG-Y65umGjG
zSrbovd|r~z&cIM$4+>tjH)YQDhK@1}3=OxH7VN$OTE(}_R~odQ`9K3>)ZVJn-rwhD
zoBOLxKKb(5+qZW+t}<nu+P3_ert<!fePP|dj%<U(w*@2P-BvN3hy_QExcJUC^L>47
zZSeVdwx&r(IF>G7e*MRs0|y+IE?Huc$Id5{aiojUBvrdS{LDSO1-Tt-<o^5uxyM{?
zo!$b=*_Or2^7eiWgOsT0=VWSsKAk?xx_sR-pP4~#i+$84D=8~WZ}wsEb=i4#-SzrA
z+Oy9F<g8eq<gAsaSQj?aNA2&9)tjoTy~V@17#JE}30#O>vSmw1_VsmFudED~y?rKa
zvyhO`maMC*q~=ebKY#s>9Tqp=mVJGB+5NoD=bl|<Z=)XnQ&m;f($p-xbN=kvwey7;
z9+ZI^W#@J81*U>VXp)?2wyN94PcrHG2r41-I_>)OCQY2EIQ#6fjt-9HZv_Pfy?lMw
zezSZ&r?~Igxw)(T?f-`S|9$^|%J$o)6(160tx8rLyFSw}xvi<GNoM<u88fPE`5E3N
ze|*+;U)DCc>FWO6jzvvc&sLn!*Sx-MO3c(-Z?Z(+EarRoZ=H@X0|SHU0@l5pk|r4w
z*m$L$?69x+kN}#mT)+3*ta&w`PG-$FE`1fEs-|Z3?)kH4LYEph2h@4)T+IFZbNKgw
z;~fm1L4`AYK{KrhVGH?E#aGtP7MSUy7CsFWSW*HPVpVi>e8jq4Gcz;&ZePB1Da-cQ
zoy(Vl_f~%gEwIkM1F3T3(nGWvzHhH_-{qBCww3pDahBruZ(CN>?%03pauz7-etYrg
zol@B9(AwYMu3lQ|EqnV-+3sn2v9GRp)ft)1_ML5(8x#@};yjgeLFbA$_v{vBtLQ#q
zU|{eS(S4yeWyXvZA0HoI9lyWs)&2eZL7Uu~nwXBSdH48OZ}94{wI<2OcswtG7Tz~9
zv#+ZC{q5AbbD+k@dS-UM7t^Bi3ZG^&6l{ECG<#a^`iOajF4sGP-|#XpFqFGGlrJ<)
zJ_g==_GaU8P=7OSZ&l{UM@RcQA5WSzDaGvn#gh+TTwJ{C>$T{qvu3Rl(~Vk@e}CVv
zpU>xCUFt0!K9`BX{KMyYJfc%>iAcU!J8x|>WC_%(6DbmH4__%=s3<Bf78Ma$GG~qq
zD;wLcT`~+`9~|U2f7N`SFZ0g=3*Y^FW`j#GUzVE({u;2!GR)dgG&}Zw?&>%9;&yYY
zF)}bbm<P(zEudNCNY<rX-@b<Y-=+vDXVO9Y?0s*TM{A|!`@fJ;<mcXhiIIU}meIdi
z%nfR%40zlf<UPNaF)%c!^*8V__+EM6UNmJ=uulAHS7+;%ox5Zh7#QX#GFdU01WsN1
zTzrnO9;{G404~RcQzY8N^4D>^P~HVDS|8*%IIuUUfqJdq(nL;V8NO~Rjsp3Kf#C`>
z$QZ||cW<|E;SX)tz8qY(JJf>8IN>cjioTWY+Rn_tFvk?+nIke*Q7g{27pIzquBqn(
zResE%qHRY0wk4jE&gBX(3aHly7n%*foIn;nn|<o$?cJ9^W!{4vu*egSxXx=bCng>F
zl*PcnkS-0fT+w^mvXcf&<kmJ{(YkT`>Qe>=h6Qqh7Z?|C{>%#QQi0gt3$ouO<=ONX
z&ut|qE%B~)*$FQ5ub6|35WJt`Ir;8{U0b<AVphxtM~s85@&aZD;g3(3rFdRi@$RDD
zI?xI+UQmtf>)=+`8TR(%6qVjre&Ar<EC}*P!_nEnsrF{o)j2EHL(E_X2g8S_??pFn
zzg)$~z)<i_X#sP?Z0kwki{ihBmgMhm0o88>*TACDdzLP$u0DUwv5RwIwKW67fty_o
zd<=Q-PT%i8d+*#WP!W@E%4Ef`W8;@SNuUKYLU(U7GBBJgV<}-MNWP!rH#N}%RBG&E
zg@~tt#p7jW@BJGTbsnU$3M3Nza_7;gz55qc|2}$GIW`8AA*7@jWf|^Vxqf`t_V0a5
zL3TS7J2<d2n16k`@!i|SFYfU%FkH|PxWIUTcj;5H?6Zxh-n|7aJ#2UeN{M+dUi)eq
z@vQXE*$~D0dj~jqnJ|GIx=`O-_VvAc%D2F402gconQ(T=Gm-4G{}xYL;ynWtvkPu2
zEMRVs-gujL@1~mVDl7A2xLsMn$!@{U4+W)+3=LeZOP~Jw%`W}*>$4D0Gc7qbMiA6|
z69zfjp{Q(PUgdA|T;G#ZmmgiSeEo}c+Ev@jN+TH<7&2T~N*FAdH}AW>Qg&q}sN?@>
z%az<K7j`XKyzA}$?5Pvu7X{d_j@!L*{jaT@E1!FTa@{+Q7YqUJ(UXf!U+<2cw1k&^
zb7`Ke)7M3%?Ix)g)uv9neO1%QA;0ha@@u*^uXl$qFff=f`UtTzl$!j$k&-v5#9;M@
zE%UOsOaFR){r1!!2i>MlyImSMoujtoc)}q|1_lRXH*PD2D+fVyFt2VzJq2lav1Q)>
zQ=jh6+MLv{lKF}C<FDVY&rjr-UCr(V`5_AAAole-x0Mf1IHPtl&x5^xUL5x?e$cQL
zs2@97KjXfDDiZ_4g3g8AySvZMvyINZwPoV=?c3RSBo<s<9e({)C1?YOkJ@Cne!06F
zSeaHlN(ftMce5a3l1W~G^zm(%Cr*jn`2WF=3sq}DiNRMKUj;t@OXctH@4*`Y5)ZeT
z1}$aiGfzJ%ck@kKTPerorM8PyJlp%O?dz^CoaqZ1GFx&(;Mg2y28IPpjCWhp&dpgF
zoxe9!SlusVx?ZeF`Z<}OzkX%CyR*~O?s(ef%r`eS&a$iBb!%((^^U7d6P8?=lcf9S
z?F5yV>$q=eKZ`ivcTzKG=GnN_$ud^gWbfTS?W|aLtxZz|RA2Bh-gWiz_TF0dHcC}R
z#bVpcvuUa-Dj8Q-g__#E?^K@`P*JfXZ{N>n$25cwaK`;lT<?`z_BCI45#w8@tC#M3
zhw*QI(>7&NY|QFp8H;OsZOx8^s|gDqW&}BW@{yg`yVy@pP5pJg{@-F=X|pT8O1-?i
zLV|)G?U+A%cJ%TpMi0rKs<Zb0YEXH(&OX%VO`6&4E4y0zzsYG{IJmgAi`TYv`qw3E
z>_M$GhJa47FKSBq`u^s3OD5l|c+A_{+Isa@rE%JsfPjDrkHv(AFE@lTrElM<{e9(*
zGwbH9Z8lkRSl+(f@c1>o^>1AYr|P~a+{*D8lv?aE9Lpze*kBN;BerzuQp;R_Z|`8x
z+O)sl-`rgM_xJbLn@;N$=WM>MeQbACMMcK(KH0AR{`H^stPEbRWoUR2G%zuJ0^b!!
zzr>mUcCT9WwdND++qt`z9dG&JvFdM7cAr=Y*u{O$4&{QYuddox`#bC9rKNpKFF~sH
zRlDEqT7Bx@T&vPc%Y0|g+4lFb{QrQsxOxBH|NpoC*oC)GpQ`HXudiA!pOFWewD_s|
z^p%L@f|hgVoP+;mIIh3z^RA?joq@rjlml1o4`~hN-Lcr3b=Au_?aYMj+qds3ef{jq
zp9c>TJXdX!b@+QqU-Npj{rkxw7pwBjto`4FSx&Cbjk@r*?6Ba4$4-A<?F<6t+EkVj
zanRO7&`MI!GKZBbSAupBiRs6^nPH))sp*+!e3^w|-M4r8?4Xq#Wr=mWmQ;sKJ?_7p
z)mC?X#L}l_k!RFgS*x#CuLTACE{+#^W!GcNLvJsQodiCX<@veU=Ij4)Gd(zxX8z?^
zzv$ZM<#XiqyLZ0Ylrc5=g7GKskdUm&{a1q?f1K3kD{x)whWk7*y9Q<kh6D0HUJEez
zC52bbGy3HHBztQ{$@{RY?}VQ`Udgb&>%2Ap+(om0H|nGnfI{jk#|wr6q1Tf)se0N*
zvtPgQDA#D_m0ve3zBs%4-#fc^6RT|-=OfU<cTj;m=SW*J<KFU|51!w<GXr-oz4mX0
zLLaCGVn_#-l$~GM`@JSr9c7tcvj19N=c5;oK$Y|^P#=O9G##UD8$apF>k!=+$JM&5
zull_6oXL9pn>uKH`~gWuS%$0+zZc3#wt<rQA!fgQDzlPTZk=3GcwSfU^2I|&o<T7h
z*$A~8>{c$Su6}<*Z26t-n_T~-#NN?X@8JHr`PBXHP(4sV+|$6vV7g+?!M`Do&oAu?
zEAu?1s~7GU7`gR#NLx9$QeR-jQo^vJEm0z>-XQwY%JhQiD)##Pt-m5wuioi8$5IID
z=}0nJF@!u$2%ETx%PV#3jH{+r8^X^2OxFD(#q?{V-n)%kKrUJ!1v1F(>8GsNMa)m1
zCa%0%ER((Py20e?)YDdspyldUL1AF(m;R?x^|(u+>DC45p*kCn@w{VTU|7M&D9f;_
z(CYDd?NGg~WiK+WihVJf9eeWT(McJO@EB@(WHfu5+H;Rdzpf;8ZuK%$oEm#)^|l4-
z>Eb8T%%40>46A4e|0@D&dT$1eO89M1wG2-+1I^nW|CG74u!ehAqJ&?+ti9u$n>nlY
zCowWG6kJnSz<hx7ntG^@z;$i42({q1hf7K}^!uo-TybVu&!3>>$N3CE-8lxo#)dfd
z*VoowUgkSn=DYphFTtR#;{5i1Hh6h^`+xFb@IA<O{Kw~8<@wV=tD`11-kY?_PvrTG
zgL6$m9q9mOrYhMxb-&+&2C%@J&_P>tpS*s(x|Lh})j@XolEPAkb0JT8-tF~IX8Z0n
z_3j;inf5w`&o_A9L4uc!<%`vgoyE(mzP?iR1RduAIs*rEy2a;bXD_?U*FL%Rh+)pe
zFME<?&*@D%sq%G&vqOGu&Y~T{=|%ol+sv$VUx31ylW{K}&*2Go^QNoRe!p7|+DWM5
zX=rG8>cokH)J%qR7uu57K9@br{N(-9!dvIpuVcMyz@sXg?by$o$P3y+5z^lf$Nu-%
z*UQ2FwoA+JRl0)~w66$Utm66m+uL9lmzM6;%mr$%HvRkgF{!UK(Y}tSD$Q{vmshUQ
z%#x_hhd^~$rQ(9?2iHVy-ja9M>Tc=v*s3QJ-LLGct@gasE?>9eU^Dyf9aaqKi;CF3
zTd{u;o0`4w|HY%WFC((6QWk9AW2>7UrT`ihi&}oaUij6;#qF1M-c6lX|F1Gq=g&u8
zlTfqW3-9wk^|1OYc}F`GmNrrhXHS`OQYaggm`$TDyiPbf%k<aR>+!2|?(ZmEY?gQD
zMN=2UmG-%rm0CJmPu@T6{89Js>yn-hhy2PYAFb`uUT1ZusmOwc%NQ1jal9xssQi>7
zYg4h|)Hl$^<4;dc>aE~#s9g~gw>I5pf=*iS+`dG^*;ll7Y?@u|YOwyzBe~<?RBpk^
zR3)2pcUNfbuP-Y%?Ja+Q@7Au;)!bT40rE1B&nLgSws-HO<V6<l4)$iv)f&b-K%te!
z^2O@Ir%ypqQBhOhYHDhJ`j8M|e<5V<OP+UAqr&yJmgVS{>;kJa<9Jc};ko^P@Np)7
zetpT&Wo6ss>V71~mRxkzTf^fJ`&49VW_i^<m6hQ>A0%b9)s`&k{=IOGljzi2Q|5&$
zd3Gs-^6>+ejCX6B#p5a*pU*9y_vq21E4$0{Jwe+TL<9vFmfx#9e)KBiithXTsotwH
zp7MSDZnZf1=$0SJx?fIxy#QL!{N{C~+i_M<%e%KBjvX|h`hNfab)YTn(&l+n`uqFU
z{N}89y?%e(ym@~Z`Al4=3Llx9`Mu0DtM*RRi{ttm(~Pd>J_2n9GT~?WV)g0s=gYa<
z?>>8F#BgOY@6^ZV&u^&S{pGkyVP@sgTcvAm+kk`VK?ui-tzT}Y&wu%-TYuW}<;(4U
zzu7$b-S6r9|16!jiYY*z=WxN92luqUok=U0z9hhYb&ZX#me%IXtM`07>s!h|y&W}{
zFHxYW2+(mha&qe~E_T1ZWB$pMNoUit_Z(uFv+?fBpH@*3YOC8L=PqVC_oG5>vZtY<
z-L09^)?fXt#Z(o;;k^b_;y&Scu{CA;?bUy!94>!9J1xZX(t{rXm6Z{%p3X4Tt=Vd<
z8-FxabiY2NRtQd<x%F~LSlBXd{XGGoQ<q9L8D5>dK3kjV*NJr#zm}a5-o0fr8|Yj(
zzt)DjSvIz|Uw^;fzk3%C!&eo}%>mJqic`%z#Y;O^Y=0bf?unkKT-xMCO?zIyy#Y$-
z)`AyaKR70xf2IE4XZi5yj0@yK%JSten0@KFZf}3@qwb5f(LYWcKD0g>)RC1*zVLd%
zo;@+e&(B>2?T_47`}@@S^Oc{L&ooZowZe#DIcNdiq$LyYzFJ?sJH(Lf?nhH_|0sYP
zR4nbO-2CgyOJz?`;nuyHF(5p1=h3L?Q*=G00{b7ISJhN|(y#Z<M}VDyq2ZRog6ju2
zrJjz7WooDjxVd+4dCmr<*OfgL4cnJD)!I&tjW`C*rz=F+T(2Em;yHQB^y%TL)sBvh
zYQD3cG<7j}OiGl9+Ntfl^JwbUTlF6|-n^~9?)kg7Q*X6~vV1Pn<pdRT7N$&9vU~o1
zyRD_AwQAGe6)Qac|9xK{8z9Kw`76zPR^FfU>vvV&lag#Jyi~iH%j@gweL9|(POJ;$
z4LUD8z5Mc(ZtKj7<I{MHYCu5~`{GfZR8etpsze*lVNg44&))wpgt9l^{BkusUem_r
zPOB)>+1uZ5Usmx1ErvCZzT{{1@%E&(D>vQbE$!qA$zkple1A>L<8BnI>4}Y?O6Qy6
zg86L^9wf~4QJd+bR{G|~##4{gCVO69?qBYh%AoS{x$Vrp_48x*#EO^kO;6`NJ=?Sd
z)I(-?P;&9;UlX0^Z9d-9^`<6D{Q7X1A9Ub{b)I(kx{#otNph7>CSJCBhHcbQFKUa&
z-(O#Y3ko*;`Sa(>>TvzNpU+vRS4AC&d<Yu5lGUBmeiM{4zNL!1_<Pm!>bu#|G0Udz
zHdeX%A3RRr5PPwyPO6PxUeEK^_Wb!t$;qikGta%K;1<&fNJ?7t?99x?jLd8q*VaTz
z878%qy}1#X8~$^`1Od<-?W-#*C$9-x4Lbe_)bxD#@S&7p(h<q4OeUqP`?k*e-ottA
zMcl=zFPC*^eZ8v}dvW>JKYQoYU)8-mZ=KQXg+Hd%y%G?7&{f39z|f$T;kdqe#}13N
zF*}3I?^P^j<Cj}=xSij7(!I*(vTI{@Zn}1SCeoN^?5>iFb1aKN746RA=V815|9n1w
zz4p2tPN2E3nwmYJ!|%SnxcGQWoUldo-#yD0xz-fatXTJ2esx)5-6<W7tFN`ch1~r9
z+E?`M?Mz!x0=5QKU+(g?O9YkOP8>h(4jP*E^z2m8R96R$dHSsF3yg|dwKe;?mZIXp
zjc0Srv=tQ{SBI^&dit67@P$iDy|*SG=L0PgHhmW!6tw8hotWj)1=|dH4jcZf38-0N
zqnl=Hv1zi(%U=2Kn!4L>Y|IZ_>3h0-nfD~Uch?LygT{dvN*x{6H!q)8<yHLb%th(^
zJ&OMR{>Q&5I=5xq-Br5P$_;fYQv-Ay%=4#Do}9RUKR#@AC}^{}xBlKG-FmxL9Q&Jo
zX2!w=3lxs)s;Mmt`^xBha-K4)*Lm%UR}ZhT+o-GJsl3Z=MScvo=+v8E-m6;6{$6)u
z=DJ5lyQ8*$2)zEnA$%$i0|P?{$UU>7);2daO_@9Q?t=BQ0s;$orOh(V&N2lZfiQK(
zj1^B$Prtt7|CN=&QzuV;yaQBuI=*I(NxXWm>)Pji8+D(&&72sran+&2(?hl_s{e4?
z@vYD4$zP7Yd8@qJZQe&4AqEBphgzit*Bw(*mXy7}7n}O~)Ku-O>tc6D$tt4`v5H7W
ze%=wW`R1B$Z*PM(4uY3!uU@||$~f)Jio$w*eSKCoHmkc0o6oL#d~Ck_`uvy<x3b<T
zcPGsITeaKkRBqI+w-MjJz3Z_DMRqkPve$0Ex2pQxPIu5QFdwzalO|98`fm4o@D_gA
z@;ia6lOxuKfo5&Jr|XsGY`*<AYFEihHeRWa*HLfEb|>@LT#&i>wk?_M`@y%)t8M;0
zI8pp05wu>S<m{ia<)6GxiOM}#de2Z-3Oo+-U`a&UTE1mIGnIDVT?eWaT3cJK-tYPR
z<muC;pdIk#_iM}7_D{w-rVGki;91kHSyxY8bh1l6#`E>{b$`&VcTLTU;qkTL2HX05
zzp|R=s9z9y<8d0)rT7`8`|6_Aj#H1{M*Vu-#<&r*+o9@h``$@Q1Y?y`&A^i-4DVVS
z;@H(Dd-mIW>X55=!1(sw-sqn{f4;i3wENHBzy7Z+l8^OVoT%*n<wkP9rm1P@-m0&c
z9v$tL6+W)2s(Lc#-{Q&6j*g9>sdHJY5)V5&J5Y%LI=(OS%nZY?Rdo)*L6+6Ocjh`h
z$y~WF_`-d?Hw*f&$ptyC^c6Y{n&&!TJCo<IK}@Vmuwmw<B~fp~FIzo(r(kX#?&#=v
z>g35p8>@3%+xP5syPxZQ{Ytm>%Wn@qn!fvf=~C>k>yzekefxSPzhC9V;x+?*28IhU
z6L|0N>%Lf?cyCW-S9f>t`+IwjK9_2EJ@xjRU#0oemUMkR@Bi%D=UbZVm&82Z#;FJz
zcsZaec)|ZLXde3ey=s5ZNts7Fg-ajznro@39J!jxcowvq=;_oc+6Qhs$1ZO6+qHiC
za##NNOFJiOfu^tg<j;C6PyF`grkAJZ#b$oH6`<n>Kua*b-OB!Y*8Kj3MSMFPp8ojU
z_|_>k@sc^HXI;qh<uWMJK@BC)R?KT_qpxnw4!>LZd~Vj|Wxbz1e~yinyRha@P0ruX
zj&J?$fkputGPpQiq#wvJJDp;5(tzh?j#>D0W)tnDPs{#p`@dqztLvaN%`i)f=~t$f
zrsl=!_j{Lj>+SN8wJy7|%y;&ePp9>@H8o#unI6d;P%h1%x_ZT#WdZi9t=iZj!;ueq
z3_b5;?)ZGp`s=^n@3qy`+$O8}UILl2VS_>VbXF5@`}hBn|LmLzp2)G_7P#Plc#dT;
zXzMR%H~G;n(Ws3{u3LZWMr~O!QJZ;|{P7>3gTXUz3?`y1UoIQu-LaVIv*iloy0DnE
z%fFW7cwR~X50sn}0?Fpzvk9Nh;Be^{vwLXPW^JPj(Yc5JfF~mv0`zC7@HtE_uK8Y8
zwH34nImgj~{lb$!dzL3PSsa_s%)kI1`o6H_$@D3&zdKt_l7Nhyb2G|1Ox~0hKP{)s
z*-+6g>+YfJ;Gu7ZuWb!{3r-sFIIsV|A$*d_mf9KMA$f*Z;vgLx3*IWOy)|1i-+%Kv
ziF@~;TDlte796zMUa|Y%m-SCiX=&I#PELI?{l!cD?VgiB+t+RcfIHj|-Z(k1Uq}Eg
z!Q1Q3_2uw6<@B00>UQ^6%~@nr+R1h0tn=+_&ZP_t3=P$a3z#zmSXG}|O$t}?yj5cH
zyr*w#j{7d?tn#;2pqc8`+@LXh`G2QaFDOjf#3dU)-_t=~L-V@5-8~)8Std)??%K|0
z_4pQO_jtLB1N#L9zLbB07q+WCKb2?xs$|Yg@3L0~WfP-Vf4|t*ZFMcY5@hjhu*K{~
z+%oacr%%bgU-&)y^>^o+Rj-%VPFk{j-3v8){r;-e{Gb65bCwbY-bV)tWACVDe$Ff_
z*57>7^yy;XP12X*tK42U6(4%N*@l6E;lNvg3ydZmr^LP}$oqRPm%s3Q-JRuYHFt7-
z`gkRIdg;pVzrYT3202jc_=5g7QkpNnsC9Xt(z&vK|7}grs+5rQ)}Pf^>l;Dto2|Hj
zc}4lsr)8??;+~V-zIIGf$+e96-T5dkcBA{z)H2>I@m1Nsw&iWG0&Tl5cXD74cn{j`
zd3Qq1S~1=({;5|Fuh}!{lh)%)*`FSRCa`ax%T}L!^2&szp!w>&<_5k6_d?k|-Vlhd
zj5_m5MpR>E+1)#njkrr+WK6U<qxLml`qF|@kkWT8Af>V=R?JY))4LKr>B`<vSBLvS
z-3|Xf&*Sk7T0bLUS}17Ne>2z#TG1CCJN;Ghyc^Bxxhj_Xmw&a9s%P`<cGCbzbSr^;
zHPv`W__hV*)?ydT<<Ct_ykZkLCs87*bPa25K}`m8I41)G!wzvKtA?uZoorPx9<JO~
z?H?E-b<>`TP1k;@B7F3%&*^Ztx|N`Xs<P5|BoID|yZNS#FD+DOW7y56ysrkUgF_mY
zOfE0}pIPqu`X{L5eVUm005p2f@EzpZ^+Cnmp;?<xUrq`NnXo;%dV5rl^Z%O{_s#`p
z?u<H)7YwgLQ@?+^rd#tg@#o`kpQkQ`(R|Ote(x!o9lQVC*-0wAndg5LgNicH%=p(=
zM>rp8$=Xk@E)_nSYc*%_o=HWlRn;r9O9BpygO2~+Aqn!-Q;rvk=NR`ox9qpkReLHj
z_2q?&i?q)F{(0k5R`Vn0RKwX)`vO48>8!v7#w*&#dEOnr=CJ<B)5MFG6HWa^jzjDN
zErI#EYMI`<NyX;hAB#@E`Q`S%#n3GOtQ!<#5^WD7p6}V!5NA30_cLLpU%QyyK`V|c
z8m2#r{$*aX5R}NILD6A8lXdASea-17LkzorckYrGztBHlV_ngopAUWn%=7`b5gb0d
zIIsugYwh0d&RTW)_T`|bAB&ZkpCp>S%t)Ncm;T={@Q(k~$LDVw{{zh!o)u@5b=bQy
zvuTG{oSax`&x#ou>o(1<jw_s5XCLzM>w^k#X|>=d*mCvR3hOR)3Ljm2Z@TL2y8$sP
z^0(f71Zo67Yif9xv^o8J*le@hCvV@b1vR|CzPf7q?)Ne2{EQnL60<HYY6YEg)Y8&2
z`5nK>wo_tXZcJwRlDAfyxoVxt%jI{cK0f~vG$GaStheEv6SNJ#%y%~E_`#Lw=jVai
z@Ru%MUVXRz=clK;{{4ErHRq<$+NiBwhK7c}KF|LTYIB<xJ$3lXCo}utUSHnWJ4;sy
z9}Qhse-7MP$au%`LeHf5na|y-*K2246fW9T@^X=7@iWkhzT)TS)|UMTjR%z9s{{>a
zE%lpg_42Iw{V8|zu3uW}Exog7L6zt6T|17qUAfbAPt(XjY_n%1>(W~_7U8#pKdX2)
z-x353sa@D3cmcK;A=JsKDS7tIoG>S+rm(e9p{tWc#KcyuS>vOqsA&2w7}n0dAn>W{
z^pwZ|t(Q-dFYL9|eI*dL@!C&m`*%}A1aCS2-7B@v1C+XY1uw*^C@VXg&Gr=(6!g2T
ztfVw$)~u-2$u?guICu5+m2v%4pM|vGz<8DP3eTr3g$wJ#SSzaMmE1p-_bxuqYoZEo
zVkKy)NJGQDkiVbj|Gxs-TL?O!=-$11rti#4Uxhq<^28#KonJ2Jh}Ok9SJK$Nd)@8#
z6-nPL4N0yH9+TQvb7!nObm&mZ#)v7?rajwYtE=lfapFYv+IQUgdlrCBV*T}Q_xmf0
z-TNoMdkvl$*u}T_q3j8(pHeT5zb;$D8aqiRZSuL?;LirUpka=k3s~>Et_fTH>)YGc
zCI_RRw&XXn@viz<dEWN>9MB10x^a6V{{H%!-4oOpWjUen73U+Z(oU`^M{_2re9e(9
zIi>dz)D$mva427R<cQ1OkH@5|e!tzmHTSj|=+t4*@k&QJg;hNtAM3pgS|IY<{`Z^B
zRbMW;o0h+e*_wITZRN_9pp|%ab^BKPF&3=Iam*Jkd)4snZD!!_X`;eM4Z*o&g@brG
zfq8|`^79w&`SohGmagvAtKso)OXg%=UiR|iaryi+mm0PHMI^nw0;-e#JTRN}{j*){
z<>{T=pvu70NU|+)$1V<&dGM`v?v0I&>-5;uJ0BU%wzA(A(WbceuRD|XGDgs>Bxn(`
zcsWD*n+q46B-@_6f4We>el>Vrnqig-lU0N9w*NagANi<FhO`f6se#1G6jm!OxSrNz
z0jkulDlcHRNM<+EPAgdd7CfHLuu2Im)OYs5J+&!fpkdVo>o{I8oQt>w8chey2{SA>
z2NLkud3Ig%ai*#mjo3S&e7_5%QCZXSsMdDrO;s~QMPIxHuWo8k>u%tC(DTOL%rVA?
zTYlzQosFOi6t0Le${vvXncQRnZh0`IH-S`HOp8bcJMBSF#g1JJ4Da&Yv*z*1>V}kT
z=wH-a3p$Ws^2$Zl*5CnShrbF7m@QV$&8*z@X6aO|tv6+F|JiFJBKv0Tiv;;xn;E*D
zVD~@p5xl^7<wTnKmkaLuS6@jBpRs@Xo3@2(nu4BApOT#(wXba1Q|oWJ_b=UVoeHWr
zedQTt9eV%m&f0%XJ@xec)6Vnn+qwig+G%NR)wEphaq_0<_QN-iW(mJqJwqT0l-V1M
z9Ua&M);ha$fBj=_t>PK1_io7w1$oa&{Z|+LnD(!0`B6~mx+46{BfbS(hYh}*I6gZx
zA}jd6nc}X6^=nppgDgD<nz;|H+4OKj*c%(A)$NhS>t5+?zWL<UBv5(3S&mUQL+zTX
z?9~&I_AlPc=Y)hgX7>J0eK!3?=lShkliuyzz#MxM6bo*LBuW-Y{-~>-^{%FTZrABI
zpnX$MtuLAW{j^T%s_m(b^D39VoS%R7;vu8bJ}%Ho^5Vdwpj`9WncJ#h+V2P3n|FQP
zpB);Lc74LPg;wfH4l`Mo?#ljGb@$)7sHIQEtiS!OtDXgK<-Pl~<%2fJ{TVu27BHu4
z9ya)L>XU8X%p2?9T=qI2ch$K~@oe0-==D1H*XD0u1{%_|wOZA!_fBqQea)W8NhPN@
zl)buWb))qvs3&r^|D*xW;j69-K0bN&OiM$<!_v}{i-)JA)G|Y%Z1+OYPPj_@_tRDG
z+U=RUtmn;>D|2hBv)0M|`cR&0Gey}mOY*@|<x15q-j7$!zt2;bUwu?`DQJ{>BBW2{
zAiTMvxJ$_X#)`0mcCRFb*UU3K`0A=v#B&$66R#4(Ja_tb%I92}zJIyLB)Pp?y7T02
zZ97$H<;wPIu{B%vJOAHb*5{YM@`&qH$vigoE_lIg1NVidduKjq&irjLBXq@m{--Y{
zy_*=SmbKb*=h=1K>FX{uo;jq|A8C6AwA;C8_CnwG;wPDgFXOL&12sVBf7<cei0e!J
z{kI|8!yI>|N`NP$P1r%Xn%nAu5@<<i*c+Qg>U!^26*k|x^2#zk*Y9fM8K<ouzFx6k
z|7xLbd-0Zs0exn2XTtmfXI@iJ^`7)+jo99@V0+%T0%5BkZ{z)ZdClvp>mPkUS^Ww-
zlhp&I-IqVBdg|-lUAy}CwrT6XKC$m>OMEdkb+@XgaQl|49qT4aT{OM)@kZkrr(ByU
zYe6|{P5Krq*H4-npaYnu2wY$k-*B7v_@^|<pQ=@_m;aqr<M?~BUHZ+w63}rM^RF)Y
zsJ>;#Y#+6)oS%bd8|-h>k_4qK=H!XV2Uv}`yh8H+C;w^(?Ka!JVau_<yoWEmNv&4#
zH0SFHp0GZ>`8o%v7AO?Fz$or<c4dYwKmUiI<tG@PicWuhRkS+mPW{(?I%{nULc;iq
zw;FE(C7Fz094{KUm&kCxxOmX6D&dIbq$QR1Z_8P}3-8^uM`=&|<?UM!G4GXL4yrmC
zdb=9<b~vO8^mBc>Hs7jsb$$%@)G4>WKKj0U)tlWv3;IAgZ?osFzk7Zh(-Q@`TUl`d
zv)+U`!783zJfEhWOu6K&Z0$J-w8`POG1r`LwO=D%|BSD4(>A-Zy^tGxL{$dJIWqT~
z^gRFGZhP^kIBM0Myo_g)=Jx&lCUf$7`0QX6&+<9-pCa#p%6pSsmXZUkTE`c-v;DN#
zcK84Du8l9|=lWcm_C&_>QnvQD69!9K&ONkJH(Kx!$@GwI3*6cN8a|8$rJ6PATU1`Q
z=9g($?)K=uCOb{#Ze^D#XsH;3Ni9psf{BuC4>?{;bUbIgNj>gLpx9>5(`7Oj_i~%9
zxRaNWuv_43XYRwUAh27Mn5-_G*u>=pYFDRBI>bDOx$3roJ!s=#(9v7ZCz;H=25K)a
zh*Vg>+`FdePvwmL)7Qv_*0a3dxuJPaU+(*RS{}}~+fB1-L5VG<xq;7QlW=;0ePGIj
z?eBbFX!v^{{gkaXd8OfLP_#Wz0v#2Vkr!0d-)qmm#%_W4CE4f4U#~xWTkssH6iaIn
zD_JmcX^Ob#s-3KRg&u;+&x|;Z7Y@qaYtz7UeGCdXR<P#X+yp*{@z$11<;f?PL~c$q
zja#_FTywcZGw2Aj1(iw*c3)^><pwQ1%)GoT^X{(FS2s5=ck7W@_~px&tnd0dIw8Bu
z-hMe>|1Wsr#EHEzy05&m^1gy6wiz<+5o$VuuI~VCS6R3FomFLJrD@UD75r_98mh~a
zc&bdm6(xgAd&4`WrAwEB?g|kS5|TlzQ`Hg=YipYl9J1_@(QXw@n+g1&HBJnF8dI;u
zpf&ln)c*cfRXumoB%!HOrdWWMe_p%2vGAAH@%in)BA>p^ECeqDX1K$t`yzd(q;Z;t
zhQ^7)?#|A_QqQxu<c)Vg8vG5$OuM>UT3b&hZPd`!J)2n_6&1DDj{&*yw`P0t<1dEI
z)}UHwmcWJBBUxKlfll+>RkAX9Hl$5#nsH%4RkbW^@bZDx%HwJsi&;Ul>IXU{n%_Bf
zc6LhH*X;ovoA&ST@9b-9B9WV7@O7S|!l!fZ?z;N%>%+tCTN4kn{rvrVb?xtOde;0`
z;x;X2ohu1yW-x&EhY{L3@#4vZIdj(B+gok@ZvTfv+*J=-#Z78{6zIn9+q3bsyR5~r
z1?-oaK&j!tOw1amq`bV5`~TLM>OPT-f4a;jJ8EZu+f)w5Ze|7cJ&x>Borw~Co{u2u
z>46I<n?0F2W%Y(DYS1$UK~1RMB}IQKH>_V?vqnrDyh{AQU4aXYZj%l{T3LQg4SXh(
zPMKa<3?A)a04-2hF!7I}qs(K_IzEOFR|j?}Pp>jphvnda1g)2GQ1(2T+6ij2Gngtb
zVD4>!oM7nJ1~$s@1-Na-09we<tFV7bOy$w1lV+%ZeChBNlyf|6uT{xCc&b>h&kuEx
ztl))4w*biti<`jRErtbEEF}vBm&nXk=zI4Tl#R2T9N48A7qi+<05_x<K+VP82Jo5z
zhB;h}vKcIzH`g?Rta$JQWW|aY=D84+riu%gdlkM$h=11r?eS({2w-Nix^SS&EBBP1
z&SV#E(6nL0EWrznY+*O6-eqsqxN!UMQP7qRhJtXGk_C-_H?r;84sW?WSjX|gfq8Wd
zv+XQ!51#=Xd<JQt^)U<q(jX0?+ZI43BO4|wE@0;EfjDEUqXWBC#8Z_*@G=Dk3rUb8
zw07@sXEX8yl_?M>gv6u;r#Y@v0XOCt=EyM0W;ndusX9qTpMP^xt!*rL1tmkdn*+Pl
zg^!QV`@$O356*#{<#jfdtsE3uTb&))r2?J`WnXjzPbV^f1BKZu6;fbrb#Y*q5`dH=
zD|DHxE;Izm=)M52hGHlv0{P+1d5sI-rr3A}<#eovG}#}#1KCjd>hxNztv5~AIM1+h
z`>`9m{E{KBuYu2`A=UcQ*Sp(FDrfl4DvK1q<^bBHP~gQ<vS8lSX}7-~c^<9mDa^J#
z@b?D>@o>=c)&-gh3z)y=yz~FPMSrnF;=j0E>I<8}$(!MVi@=4(swvZ=*G!-GZjy@c
z&K;AP_SnXPS7J6SYii)Tvber(Z!GIlv#));*QV}`-U)GtF9)OS3ibVO&tE$Jn*Vgl
zq-Fn?u0E5d)&?r)1FBg{#An-9Zv&Oa`T6;v9f+&L*W11P$1ScG0y-c4?c25Q?(TlQ
zdi}nvC0>O_X1f<=&d;5w()&_;vE16`*k}Pz!K2oWe;nZd^ZE7HzFv=CKRbV4<m>C}
zufMps*tF_PM(({mJ11%%@4Od3d5P}Hom?NjU0_x|R|wv<!Z1q#boJTl?u>PyWt5<D
zTYvwbO)4kzwu45G%kS522W>-01)TxCV!;B3_?nNc$+r(QGGARCu77r>@$#_sGiT3U
zy<maD(<e_f3=B5Z`7N1wT2JRHtG>?li-(MScdzMI<I8*vX}||7N$-&74;Q}>Gj-au
zthcweg3f?P9u#<69JV%U>5(HY%X5|8dM>EXuemf~so4zk`a7$?>=6o|UU+Pq%JUZ)
zi8If2a$N9behe8~ZcyuLc&8*ODY@xpPFF`q!7VvK!G&*cZx8=H=gO5US>OGAeb;_0
z1+QED<j!&K>{;99Rj;qRzvj!1?)rVDc-gYQOS^4!dZa)tfB<urFPrc5NE(~e{wfKd
z9=LY(uPs4UaWht*ULhX4nAz0g-x~{YP&!}d=unPj1>)B8ojpA^YyYZdzPhr~@9v}G
z;$l%@;bOL_*XKvKZ7nVZjSezAFniJe8q02hRgTSUp}*%iHnUm2)%WzCrn9oJ-t*I^
zPeBO@3oe7k1{faHWjKPn^}X`;aiGcK|Ns7iu0j(P5y^OTq;t}e!pFx}g077*Om0(o
zIj{QN#Sae;e_dby*L$&BFZjI4`*pv4K?jy=`-74uLjXUhE>AlnA!}LW(r^21Mo@4t
zC~kG*_S`TOj0p;wbmp18l_igyO~mKt=dauU`^azm=fh#p?GbiAp9q(}zIHZrIVe0B
zR;aIOzISHBHAN7wGv`4cq-tQG`Ym}jF|iC%(-S2oExB_i2C{+az*H?`$5c1Sak6{%
z*mwr{`1EvZgCg_+-=|NXL_|cMyfRWanew5~b92)6+f(PvS<}$KFw<ulTm9ZUcVdnn
zIr8Gc!DcQVo}3TY58rHj%~pDJYxAxLyBuQ+?k-q&=jAO!9_Px+%C5e?X*bWjNH2Hb
zt};$Gn|*ax>1((8y9W+9O#jd1bop>;^R7p?z<Ts{<OPJkD|&mO=v~~88K6Rkfx+AX
zTrx2n;1#%ValxKFdrWd~nN(I*?%X9}q8MYoLVKO;71ry#ubLp%PfnZ!9-d&h!vOB+
zGZ46eFe3nSym6f#XI{#&9!ab3cZx*?1QvYwPykvKXI=KD;MVdTJ0d2l`-AQ<`}6m2
z>DAD1O<moyuA+x!|9+nTf6B>}OMAcHdwt_rY4a-?yP62lRKAn5^V7$V7lX>G?f2`7
zZSo}A<f`8oy1KfaJbv7rlauq-*6iur+HKYwZsvf7lt9P9{`~Pn!_F?w-|pv<zWM1h
zv)S$cd}s#UzUS)h4!T)uvcFwn)gH6!9${hAj`hi|K5)Q+m7V?dv)TD!b#-;h^H}dN
zGRpF%9c*F+ZAIOZc^R}S?dzMH!Jr$zs;YLa_HLdXu`#Jt&41pS_`092-qUmfBO_=2
z`}5p>^@0TlHlDBjes?)&Hji6e?@E`b_LMnu)@ZNabE!4-@Z2?#o88v${Wh!W%ZrCw
z=DucAx2pQGqK!vV$=~0<_V3s0pv7PRzOJud>Dn#UR|~pE2XxlhEaUXPy1!q;fByWL
zd89*7OHZ#X@oB-<^!c@6pj(*Y>VBr?-rnYW-tPAru$C{Ey!&c*lm!F?ESP6oo%QX_
z&8v2Cv$Ip)+}L<)Q)>4#oybM{`+hiq!X<C_+il0JSa)zUXDIyr{oP#6f8LuN`I~R{
zym*n3e5^-OZw=Qw9b4PG6H?YhY&>MZbNIx$b7nTSw!N*~;#S{oZp)pmcrs#Z*3{Lf
zT4tX(ecITn<VAu`<fbDMb~QT+&)a^NXggTi{L1#r1?OWI7CL9w+{xQ6J#pg1W1r94
zzrQhU?%dpp%1X)id%wpePvI~(&AOtId})cN-|cmKK6xd-xv{Y=vEc0jEze`e<?G+<
zh@YLE^7`7^HkIVFvrPMH_x<~oeK^HPvYlT(Z_BNRRa+J;cANq!=Dj>T3UZ8RpY`(a
zxUed8bxefhok%~wJ{3(B6_-g{%HBpzojTPr)_ty3X;4m%&UHbP%1<f1Ql_U~{+h5O
zFJLd|=-$7-zFr2MV*KGl!Pj?pqx1KCY<v8%_V+hUQ`4(2cX4uYU20%t-jO$J=fP%n
z&_?H5JBydAWIjID8@01&souUHNuDv*7F%z{D6bA%8&puR;n&yK&u^$JDmtcZj@<w6
zSN7uTJl8)Klrmaar>3NwNZYKbq;x2=+Qp>>wBvr!st4!Zl<j`E_j??#q>;<rU8R#%
zmM&X%Dr+m~x|YoQ9UUA?mo3}1z-vc0=pMD#*VbO1WtzRp29$N)+}#}xN)<xF!e2ie
z=FbMLYrK8`pW5oHt3EzHZoRJe)xZ0Xc4gjM<}(wt(xmj&6;H8lSF_o^pndko`ee01
z4qtjJMtSARm0bM%<yChMA9hw$RE%*y%fl$!5Iw`Dauetth^gA)yLO4}*xAg^{|Z#@
zPg3;;-R}6R=z`0Qw`HJJ2B7VJDl0)t8uwIges*T&;*XDy@7`gxV>9T!p!fTJuX}cW
z{`z&XyR#0r@$UM5uUb@A_Uu;X1@|r<mE~0eui#jEYR$>a%*?Kyo|3@Og7EUYrPDzR
zhIFI1`K*oJUY66}(-V@JxpLFy&2zs!x5~PAZC&i@FJDSPdky8}<UW1)aG^_7`^%$l
z{ZoezI|~X5#+a{A($eyZ-BnU}>)OuZbX8SV&q<GOJ@~IX=S3sC+=VHc!6kP~udl5B
z{%+UHWwT4}*M84D)+0G-M_EDE)}3xj%F3&Kmc+!xty{B3XWinowuU~oM67$>Cva;w
zTO0B?gIsL)??*D|x<AnMbeA93{{A*~`SRtLOF?HrsH&<K%9<vN$5kve%e}RPSK7=X
z50vn3Y)qDoFTVT0#kq}V<))s;=jK|Aii@jzf{aj7R-U_V_MJO1pv&Q6%+JOoCN4aa
z_S}A^u2}b*vfV1{WamWXuAAuC%%*8=9i4rBUFn<8f4|>9owq&vKm%iMoN-LM-n$Hg
zPF%gi*jTAMN1mLVY+C=X26T-k7Y9c{&qgjTu1Wp<{Y_0xpbI0k4GjYo6%|3(^V}`J
zzt*vtE%U*FM!j`%>x2XZ9?V$2VAA~g^PQZXWpD4yD_HdV-ERM8cK&O|=WP~)?n+o4
zzW&<l_50UdkFSrNlwvfq<o(|7uU0Of7u3!tdu3a0wC%4Ki@W;zKo@4lM%%=^+H_j)
z@@0SfwKr}=l)k+c3TkPBZpO&Gq~aN4y<)=r`RniQF1LR7{`Pi%X|tRF508$d<YfQK
z-ja8BB9-0yu7DQnSQbB9v1W}AC_X@A1g`GxyI-t%Ahd3tn_T+#>2XyjL3b9c30r+}
ze|`Pq>G5@irJe=KlUD8B`}JCMUu{=cSC94kJ(n{sE@}l`FBemIRMgGOtLt@FTeMZ(
zOw+>;b~%0p-A=bSdV5~m^y$+tt_)Ux{PlW#c@?O@>$m&0LQqgJ5meTJ;(BB9aX#@-
z?(&JVXJ_9jK5y%mle6aG+qbz}vag@3crwwwXVN60<90FD5!-Six7J_ok+;8R&~qzR
zdFITS$G%>V&;RlJw{&A;<FO;1!o7R;*oe>JjX4S0AGzo}Xo#PI;ecTeXoWljxZ(%3
zNf?k$v}Xj35TLdV@F`i)X<hWB19bJt(PPI>oj&b7apFYKOd#lTE6+(MQ!c5`uX%J!
ziIL%eW+iCqfT195-L6+!Q>RS>U62&EyX@_hY15W1^`5R}Y<&6ZQbvY`%Lzt2hYLyt
z85kDiZQ8u~>LOQe-S~Ymhuitrf2x>mmRs`R0ORq$J9b!ruJM@k<>m7ES--x#1l5d-
z7B4of`%_{2>4fs94=aBBxtC(}a%y<o#q{~L%buQ|uC1lz1)8&3x9`_0H%oH{hFQ{}
zWzZK|e*XRqIu{;vhmMz**D9NRX=kN8FGc6?1s_@e7j!=(XmZN-`<=yWqqo1>dOhy)
zp32W(zFhXVw*7z6UH&TQ#;RUv^Q<>FHtu@2>-Co0+uPdg4H+0_DS|q$8ISCKy-)^C
zYiem}fo_!qje5kE-z^2LMPJf2?E`3jZ?SuSnGLA?2PJ@Iesf>l__qK5Uw?kP9}6Z;
zo-BL*_v`rotH8JSoCo!Gmix=^{d7vZx44Lb!9*F<kG^muw)ASKm!ID?kas64yJ!9U
z^z_x`<^9jj&JKTEY?gOth5EdTLl>Dqmu-PIgK27NYU${dyfM$Xps@TouarrLSP27z
z6vyNvj0^>0AHIFN2D-8ewAu%hKJ#`y6$5Q67gqCGQTG4OpBmr!cCq1cm8n~gGchpq
z_Jj1EbBf(n66omIsB-f1<zTbfXJ3A_tNjJq?38tXU+vV{H~$=;&%Xc9r|!$2>wZ2J
zFMW4s=R<!6hK4k-&9gR~IpY&seAaZ9N#>-gFE180G%)=Ac00f7-OlHrqx-#A%{W;L
zIx@s4rDM;YJ)kpzL5Fng`f|w|bSo!ly<XbpNNMxDE6>i({(2?Y-?DFi!9yodP2=3o
zcXjvseXFmo4u1{0p=7n6I77o`DUeSuv`n2c<;BP2^4GWDuLIpj3n~Uc>#Uy5%Kq}<
zF#qY}$K64<Bnt@(Z>|4d2U;h$@yspr`!$Qz=T$hpzP^6?^5x5|N?$DjU84rNHf2ed
z=cBaq^Va&?{d57fpcXA!^ySUw^RI4YuP=Ql$qd<w%FvJ%YW~}rfg!^JGb@6&H8QL^
rs$s#vASF&zOB>X~A<!7drQ}EbB|X>YO6{knfK+(8`njxgN@xNADoA_C
literal 0
HcmV?d00001
diff --git a/results/Hybridation/performance.png b/results/BERT/performance.png
similarity index 100%
rename from results/Hybridation/performance.png
rename to results/BERT/performance.png
diff --git a/temp.ipynb b/temp.ipynb
deleted file mode 100644
index 3556b7d..0000000
--- a/temp.ipynb
+++ /dev/null
@@ -1,125 +0,0 @@
-{
- "cells": [
- {
- "metadata": {
- "ExecuteTime": {
- "end_time": "2024-06-11T09:21:56.804294Z",
- "start_time": "2024-06-11T09:20:21.553571Z"
- }
- },
- "cell_type": "code",
- "source": [
- "import matplotlib.pyplot as plt\n",
- "import numpy as np\n",
- "\n",
- "# Données d'exemple\n",
- "categories = ['Cat1', 'Cat2', 'Cat3', 'Cat4', 'Cat5'] # Exemples de catégories\n",
- "metrics = ['Rappel', 'Précision', 'F-mesure']\n",
- "n_metrics = len(metrics)\n",
- "\n",
- "# Générer des performances aléatoires pour le système 1\n",
- "performance_système_1 = np.random.uniform(0.6, 0.8, size=(len(categories), n_metrics))\n",
- "\n",
- "# Générer des améliorations ou dégradations aléatoires\n",
- "amélioration_dégradation = np.random.uniform(-0.1, 0.1, size=(len(categories), n_metrics))\n",
- "\n",
- "# Calculer les nouvelles performances\n",
- "performance_système_2 = performance_système_1 + amélioration_dégradation\n",
- "\n",
- "# Créer le barplot\n",
- "fig, ax = plt.subplots(figsize=(15, 8))\n",
- "\n",
- "bar_width = 0.25\n",
- "index = np.arange(len(categories))\n",
- "\n",
- "# Couleurs pour les barres en fonction de l'amélioration ou de la dégradation\n",
- "colors = [['green' if diff > 0 else 'red' for diff in cat_diff] for cat_diff in amélioration_dégradation]\n",
- "\n",
- "# Tracer les barres pour chaque métrique\n",
- "for i, metric in enumerate(metrics):\n",
- " bars1 = ax.bar(index + i * bar_width, performance_système_1[:, i], bar_width, label=f'{metric}', alpha=0.7, color='lightgray')\n",
- " bars2 = ax.bar(index + i * bar_width, amélioration_dégradation[:, i], bar_width, bottom=performance_système_1[:, i], color=[colors[j][i] for j in range(len(categories))], alpha=0.7)\n",
- "\n",
- " # Ajouter les valeurs sur les barres du système 1\n",
- " for bar1, perf1 in zip(bars1, performance_système_1[:, i]):\n",
- " height = bar1.get_height()\n",
- " ax.annotate(f'{perf1:.2f}',\n",
- " xy=(bar1.get_x() + bar1.get_width() / 2, height),\n",
- " xytext=(0, 3), # 3 points de décalage vertical\n",
- " textcoords=\"offset points\",\n",
- " ha='center', va='bottom', color='blue')\n",
- "\n",
- " # Ajouter les valeurs sur les barres du système 2\n",
- " for bar1, bar2, perf1, diff in zip(bars1, bars2, performance_système_1[:, i], amélioration_dégradation[:, i]):\n",
- " height = bar1.get_height() + bar2.get_height()\n",
- " ax.annotate(f'{perf1 + diff:.2f}',\n",
- " xy=(bar2.get_x() + bar2.get_width() / 2, height),\n",
- " xytext=(0, 3), # 3 points de décalage vertical\n",
- " textcoords=\"offset points\",\n",
- " ha='center', va='bottom', color='black')\n",
- "\n",
- "# Ajouter des labels et une légende\n",
- "ax.set_xlabel('Catégories')\n",
- "ax.set_ylabel('Performance')\n",
- "ax.set_title('Performance du Système avec Améliorations/Dégradations')\n",
- "ax.set_xticks(index + bar_width)\n",
- "ax.set_xticklabels(categories)\n",
- "ax.legend()\n",
- "\n",
- "# Afficher le graphique\n",
- "plt.show()\n"
- ],
- "id": "c1616f9091e76aaf",
- "outputs": [
- {
- "ename": "KeyboardInterrupt",
- "evalue": "",
- "output_type": "error",
- "traceback": [
- "\u001B[1;31m---------------------------------------------------------------------------\u001B[0m",
- "\u001B[1;31mKeyboardInterrupt\u001B[0m Traceback (most recent call last)",
- "Cell \u001B[1;32mIn[12], line 16\u001B[0m\n\u001B[0;32m 13\u001B[0m amélioration_dégradation \u001B[38;5;241m=\u001B[39m np\u001B[38;5;241m.\u001B[39mrandom\u001B[38;5;241m.\u001B[39muniform(\u001B[38;5;241m-\u001B[39m\u001B[38;5;241m0.1\u001B[39m, \u001B[38;5;241m0.1\u001B[39m, size\u001B[38;5;241m=\u001B[39m(\u001B[38;5;28mlen\u001B[39m(categories), n_metrics))\n\u001B[0;32m 15\u001B[0m \u001B[38;5;66;03m# Calculer les nouvelles performances\u001B[39;00m\n\u001B[1;32m---> 16\u001B[0m performance_système_2 \u001B[38;5;241m=\u001B[39m \u001B[43mperformance_système_1\u001B[49m \u001B[38;5;241m+\u001B[39m amélioration_dégradation\n\u001B[0;32m 18\u001B[0m \u001B[38;5;66;03m# Créer le barplot\u001B[39;00m\n\u001B[0;32m 19\u001B[0m fig, ax \u001B[38;5;241m=\u001B[39m plt\u001B[38;5;241m.\u001B[39msubplots(figsize\u001B[38;5;241m=\u001B[39m(\u001B[38;5;241m15\u001B[39m, \u001B[38;5;241m8\u001B[39m))\n",
- "File \u001B[1;32m_pydevd_bundle\\\\pydevd_cython_win32_311_64.pyx:1187\u001B[0m, in \u001B[0;36m_pydevd_bundle.pydevd_cython_win32_311_64.SafeCallWrapper.__call__\u001B[1;34m()\u001B[0m\n",
- "File \u001B[1;32m_pydevd_bundle\\\\pydevd_cython_win32_311_64.pyx:627\u001B[0m, in \u001B[0;36m_pydevd_bundle.pydevd_cython_win32_311_64.PyDBFrame.trace_dispatch\u001B[1;34m()\u001B[0m\n",
- "File \u001B[1;32m_pydevd_bundle\\\\pydevd_cython_win32_311_64.pyx:937\u001B[0m, in \u001B[0;36m_pydevd_bundle.pydevd_cython_win32_311_64.PyDBFrame.trace_dispatch\u001B[1;34m()\u001B[0m\n",
- "File \u001B[1;32m_pydevd_bundle\\\\pydevd_cython_win32_311_64.pyx:928\u001B[0m, in \u001B[0;36m_pydevd_bundle.pydevd_cython_win32_311_64.PyDBFrame.trace_dispatch\u001B[1;34m()\u001B[0m\n",
- "File \u001B[1;32m_pydevd_bundle\\\\pydevd_cython_win32_311_64.pyx:585\u001B[0m, in \u001B[0;36m_pydevd_bundle.pydevd_cython_win32_311_64.PyDBFrame.do_wait_suspend\u001B[1;34m()\u001B[0m\n",
- "File \u001B[1;32mC:\\Program Files\\JetBrains\\DataSpell 2023.1.2\\plugins\\python-ce\\helpers\\pydev\\pydevd.py:1185\u001B[0m, in \u001B[0;36mPyDB.do_wait_suspend\u001B[1;34m(self, thread, frame, event, arg, send_suspend_message, is_unhandled_exception)\u001B[0m\n\u001B[0;32m 1182\u001B[0m from_this_thread\u001B[38;5;241m.\u001B[39mappend(frame_id)\n\u001B[0;32m 1184\u001B[0m \u001B[38;5;28;01mwith\u001B[39;00m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39m_threads_suspended_single_notification\u001B[38;5;241m.\u001B[39mnotify_thread_suspended(thread_id, stop_reason):\n\u001B[1;32m-> 1185\u001B[0m \u001B[38;5;28;43mself\u001B[39;49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43m_do_wait_suspend\u001B[49m\u001B[43m(\u001B[49m\u001B[43mthread\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43mframe\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43mevent\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43marg\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43msuspend_type\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43mfrom_this_thread\u001B[49m\u001B[43m)\u001B[49m\n",
- "File \u001B[1;32mC:\\Program Files\\JetBrains\\DataSpell 2023.1.2\\plugins\\python-ce\\helpers\\pydev\\pydevd.py:1200\u001B[0m, in \u001B[0;36mPyDB._do_wait_suspend\u001B[1;34m(self, thread, frame, event, arg, suspend_type, from_this_thread)\u001B[0m\n\u001B[0;32m 1197\u001B[0m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39m_call_mpl_hook()\n\u001B[0;32m 1199\u001B[0m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39mprocess_internal_commands()\n\u001B[1;32m-> 1200\u001B[0m \u001B[43mtime\u001B[49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43msleep\u001B[49m\u001B[43m(\u001B[49m\u001B[38;5;241;43m0.01\u001B[39;49m\u001B[43m)\u001B[49m\n\u001B[0;32m 1202\u001B[0m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39mcancel_async_evaluation(get_current_thread_id(thread), \u001B[38;5;28mstr\u001B[39m(\u001B[38;5;28mid\u001B[39m(frame)))\n\u001B[0;32m 1204\u001B[0m \u001B[38;5;66;03m# process any stepping instructions\u001B[39;00m\n",
- "\u001B[1;31mKeyboardInterrupt\u001B[0m: "
- ]
- }
- ],
- "execution_count": 12
- },
- {
- "metadata": {},
- "cell_type": "code",
- "outputs": [],
- "execution_count": null,
- "source": " ",
- "id": "c7044d0f9235b5c"
- }
- ],
- "metadata": {
- "kernelspec": {
- "display_name": "Python 3",
- "language": "python",
- "name": "python3"
- },
- "language_info": {
- "codemirror_mode": {
- "name": "ipython",
- "version": 2
- },
- "file_extension": ".py",
- "mimetype": "text/x-python",
- "name": "python",
- "nbconvert_exporter": "python",
- "pygments_lexer": "ipython2",
- "version": "2.7.6"
- }
- },
- "nbformat": 4,
- "nbformat_minor": 5
-}
--
GitLab