fixed shadow tree tab updating bug

.vscode/settings.json

@@ -1,5 +1,5 @@
 {
-    "python.pythonPath": "/anaconda/bin/python",
+    "python.pythonPath": "C:\\ProgramData\\Anaconda3\\envs\\ww_env2\\python.exe",
     "python.testing.unittestArgs": [
         "-v",
         "-s",

TODO.md

@@ -2,20 +2,18 @@
 # TODO:
 
 ## Layout:
-- set all tabs default to False
 
 ## Plots:
 - add multiclass confusion matrix
 - individual trees: highlight selected tree
-- fix shap dependence summary going full width
 
 ### Regression plots:
 
 ## Explainers:
 
 ## Dashboard:
 - add dependence plot to importances list
-- COntributions: add div margin to index selector
+- Contributions: add div margin to index selector
 - move number of features to display
 - add option for vertical contributions?
 - reformat contributions table
@@ -26,10 +24,12 @@
 - Move pdp function to explainer_methods.py
 - add feature explanations
 
+
 ## Library level:
 - fix forever updating bug (seems shadow tree related?)
 - fix jupyter reload pdp bug
-- just add kind='tree', 'linear', 'deep', etc
+- submit pull request to dtreeviz to accept shadowtree as parameter
+- just add shap='tree', 'linear', 'deep', etc instead of separate classes
 - add long description to pypi: https://packaging.python.org/guides/making-a-pypi-friendly-readme/
 - Add tests
 - Test with lightgbm, catboost, extratrees

explainerdashboard/dashboard_tabs/model_summary_tab.py

@@ -234,7 +234,7 @@ def layout(self):
                                                 0.75: '0.75', 0.99: '0.99'}, 
                                         included=False,
                                         tooltip = {'always_visible' : True})
-                        ], style={'margin': 20}),
+                        ], style={'margin': 0}),
                     ])
                 ]),
             dbc.Row([
@@ -247,7 +247,7 @@ def layout(self):
                                                 0.75: '0.75', 0.99: '0.99'}, 
                                         included=False,
                                         tooltip = {'always_visible' : True})
-                        ], style={'margin': 20}),
+                        ], style={'margin': 0}),
                     ])
                 ]),
             dbc.Row([
@@ -280,13 +280,6 @@ def layout(self):
 
     def register_callbacks(self, app, **kwargs):
 
-        @app.callback(
-            Output('precision-cutoff', 'value'),
-            [Input('fraction-cutoff', 'value')]
-        )
-        def update_cutoff(fraction):
-            return np.round(self.explainer.cutoff_fraction(fraction), 2)
-
         @app.callback(
             [Output('bin-size-div', 'style'),
              Output('quantiles-div', 'style')],
@@ -349,8 +342,6 @@ def update_precision_graph(cutoff, percentage, pos_label, tab):
             return self.explainer.plot_confusion_matrix(
                                 cutoff=cutoff, normalized=percentage)
 
-
-
         @app.callback(
             Output('roc-auc-graph', 'figure'),
             [Input('precision-cutoff', 'value'),
@@ -369,6 +360,13 @@ def update_precision_graph(cutoff, pos_label, tab):
         def update_precision_graph(cutoff, pos_label, tab):
             return self.explainer.plot_pr_auc(cutoff=cutoff)
 
+        @app.callback(
+            Output('precision-cutoff', 'value'),
+            [Input('fraction-cutoff', 'value')]
+        )
+        def update_cutoff(fraction):
+            return np.round(self.explainer.cutoff_fraction(fraction), 2)
+
 class RegressionModelStats:
     def __init__(self, explainer, round=2, logs=False, vs_actual=False, ratio=False):
         self.explainer = explainer

explainerdashboard/dashboard_tabs/shadow_trees_tab.py

@@ -61,8 +61,10 @@ def shadow_trees_layout(explainer, round=2, **kwargs):
                     ], inline=True),
                 dcc.Store(id='tree-index-store'),
                 html.H4('(click on a prediction to see decision path)'),
-                dcc.Loading(id="loading-trees-graph", 
-                            children=[dcc.Graph(id='tree-predictions-graph')]),  
+               # dcc.Loading(id="loading-trees-graph", 
+                           # children=[
+                                dcc.Graph(id='tree-predictions-graph'),
+                                #]),  
             ], width={"size": 8, "offset": 2})
         ]), 
         dbc.Row([
@@ -77,7 +79,7 @@ def shadow_trees_layout(explainer, round=2, **kwargs):
         ]), 
         dbc.Row([
             dbc.Col([
-                 dcc.Loading(id="loading-svg-graph", 
+                 dcc.Loading(id="loading-tree-svg-graph", 
                             children=[html.Img(id='dtreeviz-svg')])
             ])
 
@@ -86,29 +88,41 @@ def shadow_trees_layout(explainer, round=2, **kwargs):
 
 
 def shadow_trees_callbacks(explainer, app, round=2, **kwargs):
-    @app.callback(
-            Output('tree-input-index', 'value'),
-            [Input('tree-index-button', 'n_clicks')],
-            [State('tabs', 'value')]
-        )
-    def update_tree_input_index(n_clicks, tab):
-        return explainer.random_index(return_str=True)
 
     @app.callback(
-        Output('tree-index-store', 'data'),
-        [Input('tree-input-index', 'value')]
-    )
-    def update_tree_index_store(index):
-        if (explainer.idxs is None 
-            and str(index).isdigit() 
-            and int(index) >= 0
-            and int(index) <= len(explainer)):
-            return int(index)
-        if (explainer.idxs is not None
-             and index in explainer.idxs):
-            return index
+        [Output('tree-basevalue', 'children'),
+         Output('tree-predictions-table', 'columns'),
+         Output('tree-predictions-table', 'data')],
+        [Input('tree-predictions-graph', 'clickData')],
+         #Input('label-store', 'data')], #this causes issues for some reason, only on this tab??
+        [State('tree-index-store', 'data'),
+         State('tabs', 'value')])
+    def display_tree_click_data(clickData, idx, tab):
+        if clickData is not None and idx is not None:
+            model = int(clickData['points'][0]['text'].split('tree no ')[1].split(':')[0]) if clickData is not None else 0
+            (baseval, prediction, shadowtree_df) = \
+                explainer.shadowtree_df_summary(model, idx, round=round)
+            columns=[{'id': c, 'name': c} 
+                        for c in  shadowtree_df.columns.tolist()]
+            baseval_str = f"Tree no {model}, Starting prediction   : {baseval}, final prediction : {prediction}"
+            print(baseval, columns, shadowtree_df)
+            return (baseval_str, columns, shadowtree_df.to_dict('records'))
         raise PreventUpdate
 
+    @app.callback(
+        Output('dtreeviz-svg', 'src'),
+        [Input('tree-predictions-graph', 'clickData'),
+         #Input('label-store', 'data')#this causes issues for some reason, only on this tab??
+         ],
+        [State('tree-index-store', 'data'),
+         State('tabs', 'value')])
+    def display_click_data(clickData, idx, tab):
+        if clickData is not None and idx is not None and explainer.graphviz_available:
+            model = int(clickData['points'][0]['text'].split('tree no ')[1].split(':')[0]) 
+            svg_encoded = explainer.decision_path_encoded(model, idx)
+            return svg_encoded
+        return ""
+
     @app.callback(
         Output('tree-predictions-graph', 'figure'),
         [Input('tree-index-store', 'data'),
@@ -118,38 +132,27 @@ def update_tree_index_store(index):
     def update_tree_graph(index, pos_label, tab):
         if index is not None:
             return explainer.plot_trees(index, round=round)
-        raise PreventUpdate
+        return {}
 
     @app.callback(
-        [Output('tree-basevalue', 'children'),
-         Output('tree-predictions-table', 'columns'),
-         Output('tree-predictions-table', 'data'),],
-        [Input('tree-predictions-graph', 'clickData'),
-         Input('tree-index-store', 'data'),
-         Input('label-store', 'data')],
-        [State('tabs', 'value')])
-    def display_click_data(clickData, idx, pos_label, tab):
-        if clickData is not None and idx is not None:
-            model = int(clickData['points'][0]['text'].split('tree no ')[1].split(':')[0]) if clickData is not None else 0
-            (baseval, prediction, 
-                    shadowtree_df) = explainer.shadowtree_df_summary(model, idx, round=round)
-            columns=[{'id': c, 'name': c} for c in  shadowtree_df.columns.tolist()]
-            baseval_str = f"Tree no {model}, Starting prediction   : {baseval}, final prediction : {prediction}"
-            return (baseval_str, columns, shadowtree_df.to_dict('records'))
-        raise PreventUpdate
-
+        Output('tree-index-store', 'data'),
+        [Input('tree-input-index', 'value')],
+        [State('tabs', 'value')]
+    )
+    def update_tree_index_store(index, tab):
+        if (explainer.idxs is None 
+            and str(index).isdigit() 
+            and int(index) >= 0
+            and int(index) <= len(explainer)):
+            return int(index)
+        if (explainer.idxs is not None and index in explainer.idxs):
+            return index
+        return None
 
     @app.callback(
-        Output('dtreeviz-svg', 'src'),
-        [Input('tree-predictions-graph', 'clickData'),
-         Input('tree-index-store', 'data'),
-         Input('label-store', 'data')],
-        [State('tabs', 'value')])
-    def display_click_data(clickData, idx, pos_label, tab):
-        if (clickData is not None 
-            and idx is not None 
-            and explainer.graphviz_available):
-            model = int(clickData['points'][0]['text'].split('tree no ')[1].split(':')[0]) 
-            svg_encoded = explainer.decision_path_encoded(model, idx)
-            return svg_encoded
-        raise PreventUpdate
+        Output('tree-input-index', 'value'),
+        [Input('tree-index-button', 'n_clicks')],
+        [State('tabs', 'value')]
+    )
+    def update_tree_input_index(n_clicks, tab):
+        return explainer.random_index(return_str=True)

explainerdashboard/dashboard_tabs/shap_dependence_tab.py

@@ -40,7 +40,7 @@ def register_callbacks(self, app):
         shap_dependence_callbacks(self.explainer, app)
 
 
-def shap_dependence_layout(explainer, n_features=10, **kwargs):
+def shap_dependence_layout(explainer, n_features=10, cats=True, **kwargs):
 
     cats_display = 'none' if explainer.cats is None else 'inline-block'
     return dbc.Container([
@@ -52,8 +52,8 @@ def shap_dependence_layout(explainer, n_features=10, **kwargs):
                     dbc.Label("Depth:"),
                     dcc.Dropdown(id='dependence-scatter-depth',
                         options = [{'label': str(i+1), 'value':i+1} 
-                                        for i in range(len(explainer.columns)-1)],
-                        value=min(n_features, len(explainer.columns)-1))],
+                                        for i in range(len(explainer.columns_ranked(cats))-1)],
+                        value=min(n_features, len(explainer.columns_ranked(cats))-1))],
                     width=3), 
                 dbc.Col([
                     dbc.FormGroup(
@@ -97,15 +97,15 @@ def shap_dependence_layout(explainer, n_features=10, **kwargs):
                     html.Label('Plot dependence for column:'),
                     dcc.Dropdown(id='dependence-col', 
                         options=[{'label': col, 'value':col} 
-                                    for col in explainer.columns_cats],
-                        value=explainer.columns_cats[0])],
+                                    for col in explainer.columns_ranked(cats)],
+                        value=explainer.columns_ranked(cats)[0])],
                     width=5), 
                 dbc.Col([
                      html.Label('Color observation by column:'),
                     dcc.Dropdown(id='dependence-color-col', 
                         options=[{'label': col, 'value':col} 
-                                    for col in explainer.columns_cats],
-                        value=explainer.columns_cats[0])],
+                                    for col in explainer.columns_ranked(cats)],
+                        value=explainer.columns_ranked(cats)[1])],
                     width=5), 
                 dbc.Col([
                     html.Label('Highlight:'),

explainerdashboard/dashboard_tabs/shap_interactions_tab.py

@@ -67,7 +67,7 @@ def shap_interactions_layout(explainer,
                 dbc.Col([
                     dbc.Label("Feature"),
                     dcc.Dropdown(id='interaction-col', 
-                        options=[{'label': col, 'value':col} 
+                        options=[{'label': col, 'value': col} 
                                     for col in explainer.columns_ranked(cats)],
                         value=explainer.columns_ranked(cats)[0])],
                     width=4), 

explainerdashboard/dashboards.py

@@ -116,7 +116,9 @@ def _insert_tabs(self):
         if self.shap_interaction:
             self.tabs.append(ShapInteractionsTab(self.explainer, **self.kwargs))
         if self.shadow_trees:
-            assert hasattr(self.explainer, 'shadow_trees')
+            assert hasattr(self.explainer, 'shadow_trees'), \
+                """the explainer object has no shadow_trees property. This tab 
+                only works with a RandomForestClassifierBunch or RandomForestRegressionBunch""" 
             self.tabs.append(ShadowTreesTab(self.explainer, **self.kwargs))
 
     def run(self, port=8050, **kwargs):