diff --git a/callbacks.py b/callbacks.py
index 06b112f9a3e78088f141fe86f96c473977e239bd..906b09c41a107b7943cb9842a8a4a293fd5a649a 100644
--- a/callbacks.py
+++ b/callbacks.py
@@ -14,6 +14,9 @@ from pages.application.RandomForest.utils import xrf
from pages.application.RandomForest.utils.xrf import *
sys.modules['xrf'] = xrf
+from sklearn.ensemble._voting import VotingClassifier
+from sklearn.ensemble import RandomForestClassifier
+
from pages.application.DecisionTree.DecisionTreeComponent import DecisionTreeComponent
from pages.application.RandomForest.RandomForestComponent import RandomForestComponent
@@ -30,12 +33,13 @@ def register_callbacks(app):
# For course directory
main_course = dcc.Tabs(children=[
dcc.Tab(label='Data format', children=[]),
- dcc.Tab(label='Course Decision Tree', children=[])])
+ dcc.Tab(label='Course Decision Tree', children=[]),
+ dcc.Tab(label='Course Random Forest', children=[])])
page_course = dbc.Row([main_course])
+ # For the application
models_data = open('data_retriever.json')
data = json.load(models_data)["data"]
- # For the application
names_models, dict_components, dic_solvers, dic_xtypes = extract_data(data)
# region alerts
@@ -43,7 +47,7 @@ def register_callbacks(app):
is_open=True,
color='warning',
duration=10000, ), ])
- warning_selection_pretrained_model = html.Div([dbc.Alert("You uploaded the data, now upload the pretrained model.",
+ warning_selection_pretrained_model = html.Div([dbc.Alert("Upload the pretrained model.",
is_open=True,
color='warning',
duration=10000, ), ])
@@ -113,7 +117,7 @@ def register_callbacks(app):
xtypes = dic_xtypes[ml_type]
xtype = [list(xtypes.keys())[0]]
return solvers, solver, xtypes, xtype
- else :
+ else:
return [], None, [], None
# endregion
@@ -130,7 +134,7 @@ def register_callbacks(app):
ihm_id = ctx.triggered_id
if ihm_id == 'ml_model_choice':
return None
- else :
+ else:
return filename
# endregion
@@ -178,22 +182,21 @@ def register_callbacks(app):
@app.callback(Output('instance_filename', 'children'),
Input('ml_model_choice', 'value'),
Input('ml_pretrained_model_choice', 'contents'),
- Input('ml_instance_choice', 'contents'),
- State('ml_instance_choice', 'filename'),
+ Input('ml_instance_choice', 'filename'),
prevent_initial_call=True
)
- def select_instance(ml_type, model, instance, filename):
+ def select_instance(ml_type, model, filename):
ctx = dash.callback_context
if ctx.triggered:
ihm_id = ctx.triggered_id
if ihm_id == 'ml_model_choice' or ihm_id == "ml_pretrained_model_choice":
return None
- else :
+ else:
return filename
# endregion
- # region draw
+ # region main
@app.callback(
Output('graph', 'children'),
Output('explanation', 'children'),
@@ -215,24 +218,31 @@ def register_callbacks(app):
Input('solver_sat', 'value'),
Input('choice_tree', 'value'),
prevent_initial_call=True)
- def draw_model(ml_type, pretrained_model, model_filename, need_data, data, instance, instance_filename, click, model_info_filename,
+ def draw_model(ml_type, pretrained_model, model_filename, need_data, data, instance, instance_filename, click,
+ model_info_filename,
expl_choice, cont_expl_choice, enum, xtype, solver, id_tree):
ctx = dash.callback_context
if ctx.triggered:
ihm_id = ctx.triggered_id
- if ihm_id == "ml_model_choice" :
+ # selecting a machine learning model type initialize the component
+ if ihm_id == "ml_model_choice":
return reinit, None, {}, {}
+ # uploading a model
elif ihm_id == "ml_pretrained_model_choice":
return init_network, None, {}, {}
- else :
+ else:
+ # construction of the component base from pretrained model
+ # catching exception if the construction of the component fails with try/except/else
try:
- pretrained_model = parse_contents_graph(pretrained_model, model_filename)
if ml_type is None:
return warning_selection_model, None, {}, {}
+ elif pretrained_model is None :
+ return warning_selection_pretrained_model, None, {}, {}
else:
+ pretrained_model = parse_contents_graph(pretrained_model, model_filename)
component_class = dict_components[ml_type]
component_class = globals()[component_class]
if not need_data:
@@ -241,40 +251,33 @@ def register_callbacks(app):
component = component_class(pretrained_model, info=data, type_info=model_info_filename)
else:
return warning_selection_data, None, {}, {}
-
except:
return alert_network, None, {}, {}
- else :
+ else:
+ # plotting model by clicking "submit" button
if ihm_id == "submit-model":
return component.network, None, {}, {}
+ # construction of explanation
+ if instance is not None:
+ try:
+ instance = parse_contents_instance(instance, instance_filename)
+ component.update_with_explicability(instance, enum, xtype, solver)
+ # In the case of DecisionTree, plotting explanation
+ if ihm_id == 'expl_choice':
+ component.draw_explanation(instance, expl_choice)
+ # # In the case of DecisionTree, plotting cont explanation
+ elif ihm_id == 'cont_expl_choice':
+ component.draw_contrastive_explanation(instance, cont_expl_choice)
+ except:
+ return component.network, alert_explanation, {}, {}
+
# In the case of RandomForest, id of tree to choose to draw tree
- elif ihm_id == 'choice_tree':
+ if ihm_id == 'choice_tree':
component.update_plotted_tree(id_tree)
- return component.network, component.explanation, {}, {}
-
- elif instance is not None:
- instance = parse_contents_instance(instance, instance_filename)
- list_expls, list_cont_expls = component.update_with_explicability(instance, enum, xtype, solver)
- options_expls = {}
- options_cont_expls = {}
- for i in range(len(list_expls)):
- options_expls[str(list_expls[i])] = list_expls[i]
- for i in range(len(list_cont_expls)):
- options_cont_expls[str(list_cont_expls[i])] = list_cont_expls[i]
-
- if ihm_id == 'expl_choice':
- component.draw_explanation(instance, expl_choice)
- return component.network, component.explanation, options_expls, options_cont_expls
-
- # Choice of CxP to draw
- elif ihm_id == 'cont_expl_choice':
- component.draw_contrastive_explanation(instance, cont_expl_choice)
- return component.network, component.explanation, options_expls, options_cont_expls
-
- else :
- return component.network, component.explanation, options_expls, options_cont_expls
+
+ return component.network, component.explanation, component.options_expls, component.options_cont_expls
# endregion
@@ -284,17 +287,16 @@ def register_callbacks(app):
Output('explanation', 'hidden'),
Input('ml_model_choice', 'value'),
Input('ml_pretrained_model_choice', 'contents'),
- State('ml_instance_choice', 'contents'),
Input('submit-instance', 'n_clicks'),
prevent_initial_call=True
)
- def show_explanation_window(ml_type, model, instante, click):
+ def show_explanation_window(ml_type, model, click):
ctx = dash.callback_context
if ctx.triggered:
ihm_id = ctx.triggered_id
- if ihm_id == "ml_model_choice" or ihm_id == "ml_pretrained_model_choice" or ihm_id == "ml_instance_choice":
+ if ihm_id == "ml_model_choice" or ihm_id == "ml_pretrained_model_choice":
return True
- else :
+ else:
return False
# endregion
@@ -309,6 +311,24 @@ def register_callbacks(app):
def choose_tree_in_forest(ml_type):
return bool(ml_type != "RandomForest")
+ @app.callback(
+ Output("choice_tree", "max"),
+ State('ml_model_choice', 'value'),
+ Input('ml_pretrained_model_choice', 'contents'),
+ State('ml_pretrained_model_choice', 'filename'),
+ prevent_initial_call=True)
+ def adjust_slider_max(ml_type, pretrained_model, model_filename):
+ if ml_type == "RandomForest":
+ pretrained_model = parse_contents_graph(pretrained_model, model_filename)
+ if isinstance(pretrained_model, xrf.rndmforest.RF2001):
+ return int(pretrained_model.forest.n_estimators)
+ elif isinstance(pretrained_model, RandomForestClassifier):
+ return pretrained_model.n_estimators
+ elif isinstance(pretrained_model, VotingClassifier):
+ return len(pretrained_model.estimators)
+ else:
+ return 0
+
# endregion
# region decisiontree
@@ -340,4 +360,4 @@ def register_callbacks(app):
def switcher_drawing_options(bool_draw):
return not bool_draw
- # endregion
\ No newline at end of file
+ # endregion
diff --git a/data_retriever.json b/data_retriever.json
index 1607a1a1fcc43cca6c7025fa1ea37db6076e535f..e0051a0150d14c7b9315b1d50b3907f1f72ae4ad 100644
--- a/data_retriever.json
+++ b/data_retriever.json
@@ -8,21 +8,20 @@
"g3", "g4", "lgl", "mcb", "mcm", "mpl", "m22", "mc", "mgh"
],
"xtypes" : {
- "AXp": "Abductive Explanation", "CXp": "Contrastive explanation"}
- },
- {
- "ml_type" : "NaiveBayes",
- "component" : "NaiveBayesComponent",
- "solvers" : [],
- "xtypes" : {
- "AXp": "Abductive Explanation", "CXp": "Contrastive explanation"}
+ "AXp": " Abductive ", "CXp": " Contrastive "}
},
{
"ml_type" : "RandomForest",
"component" : "RandomForestComponent",
"solvers" : ["SAT"],
- "xtypes" : {"M": "Minimal explanation"}
+ "xtypes" : {"abd": " Abductive ", "con" : " Contrastive "}
+ },
+ {
+ "ml_type" : "NaiveBayes",
+ "component" : "NaiveBayesComponent",
+ "solvers" : ["SAT"],
+ "xtypes" : {"abd": " Abductive ", "con" : " Contrastive "}
}
]
-}
\ No newline at end of file
+}
diff --git a/pages/application/DecisionTree/DecisionTreeComponent.py b/pages/application/DecisionTree/DecisionTreeComponent.py
index 81f00168f609cc3bd2261bb5c4de42ae84542e18..96f8146573b3cd4394c10144ca3580da70d3478a 100644
--- a/pages/application/DecisionTree/DecisionTreeComponent.py
+++ b/pages/application/DecisionTree/DecisionTreeComponent.py
@@ -62,6 +62,8 @@ class DecisionTreeComponent:
"background-color": "transparent"})])
# init explanation
self.explanation = []
+ self.options_cont_expls = {}
+ self.options_expls = {}
def create_fvmap_inverse(self, instance):
def create_fvmap_inverse_with_info(features_names_mapping):
@@ -158,7 +160,12 @@ class DecisionTreeComponent:
"background-color": "transparent"}
)])
- return list_explanations_path, list_contrastive_explanations_path
+ self.options_expls = {}
+ self.options_cont_expls = {}
+ for i in range(len(list_explanations_path)):
+ self.options_expls[str(list_explanations_path[i])] = list_explanations_path[i]
+ for i in range(len(list_contrastive_explanations_path)):
+ self.options_cont_expls[str(list_contrastive_explanations_path[i])] = list_contrastive_explanations_path[i]
def draw_explanation(self, instance, expl):
r""" Called with the selection of an explanation to plot on the tree
diff --git a/pages/application/RandomForest/RandomForestComponent.py b/pages/application/RandomForest/RandomForestComponent.py
index 38522ba30353ae06e6a54059f5dde35e587fecdd..19704e42a8ef3dcb913c3c2d27c2459225eaf0ba 100644
--- a/pages/application/RandomForest/RandomForestComponent.py
+++ b/pages/application/RandomForest/RandomForestComponent.py
@@ -1,8 +1,13 @@
+import re
+
+import numpy
from dash import html
import dash_interactive_graphviz
from sklearn import tree
from pages.application.RandomForest.utils import xrf
from pages.application.RandomForest.utils.xrf.xforest import XRF, Dataset
+from sklearn.ensemble._voting import VotingClassifier
+from sklearn.ensemble import RandomForestClassifier
class RandomForestComponent:
@@ -15,14 +20,25 @@ class RandomForestComponent:
# creation of model
if info is not None and 'csv' in type_info:
- self.random_forest = XRF(model, self.data.feature_names, self.data.target_name)
+ if isinstance(model, xrf.rndmforest.RF2001):
+ self.random_forest = XRF(model, self.data.feature_names, self.data.target_name)
+ elif isinstance(model, RandomForestClassifier):
+ params = {'n_trees': model.n_estimators,
+ 'depth': model.max_depth}
+ cls = xrf.rndmforest.RF2001(**params)
+ train_accuracy, test_accuracy = cls.train(self.data)
+ self.random_forest = XRF(cls, self.data.feature_names, self.data.target_name)
# encoding here so not in the explanation ?
# visual
self.tree_to_plot = 0
dot_source = tree.export_graphviz(self.random_forest.cls.estimators()[self.tree_to_plot],
feature_names=self.data.feature_names, class_names=list(map(lambda cl : str(cl), self.data.target_name)),
- impurity=False, filled=False, rounded=True)
+ impurity=False, filled=True, rounded=True)
+ dot_source = re.sub(r"(samples) \= [0-9]+", '', dot_source)
+ dot_source = re.sub(r"\\nvalue \= \[[\d+|\,|\s]+\]\\n", '', dot_source)
+ dot_source = re.sub(r"\\nclass \= \d+", '', dot_source)
+
self.network = html.Div([dash_interactive_graphviz.DashInteractiveGraphviz(
dot_source=dot_source, style={"width": "50%",
"height": "80%",
@@ -31,14 +47,16 @@ class RandomForestComponent:
# init explanation
self.explanation = []
+ self.options_cont_expls = {}
+ self.options_expls = {}
- def update_with_explicability(self, instances, enum_feats=None, xtype=None, solver=None):
+ def update_with_explicability(self, instances, enum_feats=None, xtypes=None, solver=None):
r""" Called when an instance is upload or when you press the button "Submit for explanation" with advanced parameters.
Args:
instances : list - list of instance to explain
enum_feats : ghost feature
- xtype : ghost feature
- solver : ghost feature
+ xtypes : types of explanation
+ solver : solver, only SAT available for the moment
"""
instances = [list(map(lambda feature: feature[1], instance)) for instance in instances]
self.explanation = []
@@ -54,18 +72,18 @@ class RandomForestComponent:
self.explanation.append(html.Hr())
# Call explanation
- explanation_result = None
- if isinstance(self.random_forest, XRF):
- explanation_result = self.random_forest.explain(instance)
- # Creating a clean and nice text component
- for k in explanation_result.keys():
- self.explanation.append(html.H5(k))
- self.explanation.append(html.Hr())
- self.explanation.append(html.P(explanation_result[k]))
- self.explanation.append(html.Hr())
+ xtypes_trad = {"abd": " Abductive ", "con" : "Contrastive "}
+ for xtype in xtypes :
+ explanation_result = self.random_forest.explain(instance, xtype)
+ # Creating a clean and nice text component
+ for k in explanation_result.keys():
+ self.explanation.append(html.H5(xtypes_trad[xtype] + k))
+ self.explanation.append(html.Hr())
+ self.explanation.append(html.P(explanation_result[k]))
+ self.explanation.append(html.Hr())
- del self.random_forest.enc
- del self.random_forest.x
+ del self.random_forest.enc
+ del self.random_forest.x
return [], []
@@ -78,7 +96,10 @@ class RandomForestComponent:
self.tree_to_plot = tree_to_plot
dot_source = tree.export_graphviz(self.random_forest.cls.estimators()[self.tree_to_plot],
feature_names=self.data.feature_names, class_names=list(map(lambda cl : str(cl), self.data.target_name)),
- impurity=False, filled=False, rounded=True)
+ impurity=False, filled=True, rounded=True)
+ dot_source = re.sub(r"(samples) \= [0-9]+", '', dot_source)
+ dot_source = re.sub(r"\\nvalue \= \[[\d+|\,|\s]+\]\\n", '', dot_source)
+ dot_source = re.sub(r"\\nclass \= \d+", '', dot_source)
self.network = html.Div([dash_interactive_graphviz.DashInteractiveGraphviz(
dot_source=dot_source, style={"width": "50%",
diff --git a/pages/application/application.py b/pages/application/application.py
index c53d260ec6fba2e8d863efb3dc5e79325bff7bd5..ced07ee303faca2b26bae369f3669cecb3d52d1e 100644
--- a/pages/application/application.py
+++ b/pages/application/application.py
@@ -123,7 +123,7 @@ class View:
inline=True),
html.Hr()])
- self.solver = html.Div([html.Label("Choose the SAT solver : "),
+ self.solver = html.Div([html.Label("Choose the solver : "),
html.P(),
dcc.Dropdown(self.model.solvers,
id='solver_sat')])
@@ -169,9 +169,10 @@ class View:
self.tree_to_plot = html.Div(id="choosing_tree", hidden=True,
children=[html.H5("Choose a tree to plot: "),
- html.Div(children=[dcc.Slider(0, 50, 1,
- value=0,
- id='choice_tree')])])
+ html.Div(children=[dcc.Slider(0, 20, 1, marks=None,
+ tooltip={"placement": "bottom", "always_visible": True},
+ value=0,
+ id='choice_tree')])])
self.layout = dbc.Row([dbc.Col([self.sidebar],
width=3, class_name="sidebar"),