Enanchment in CF Generation

Signed-off-by: giandos200 <giando95menico@gmail.com>

Author: giandos200

dice_ml/__init__.py

@@ -1,6 +1,6 @@
 from .data import Data
-from .dice import Dice
 from .model import Model
+from .dice import Dice
 
 __all__ = ["Data",
            "Model",

dice_ml/counterfactual_explanations.py

@@ -1,12 +1,11 @@
 import json
-import os
-
 import jsonschema
+import os
 
-from dice_ml.constants import _SchemaVersions
-from dice_ml.diverse_counterfactuals import (CounterfactualExamples,
-                                             _DiverseCFV2SchemaConstants)
+from dice_ml.diverse_counterfactuals import CounterfactualExamples
 from dice_ml.utils.exception import UserConfigValidationException
+from dice_ml.diverse_counterfactuals import _DiverseCFV2SchemaConstants
+from dice_ml.constants import _SchemaVersions
 
 
 class _CommonSchemaConstants:

dice_ml/data_interfaces/private_data_interface.py

@@ -1,11 +1,10 @@
 """Module containing meta data information about private data."""
 
-import collections
-import logging
 import sys
-
-import numpy as np
 import pandas as pd
+import numpy as np
+import collections
+import logging
 
 from dice_ml.data_interfaces.base_data_interface import _BaseData
 

dice_ml/data_interfaces/public_data_interface.py

@@ -1,15 +1,13 @@
 """Module containing all required information about the interface between raw (or transformed)
 public data and DiCE explainers."""
 
+import pandas as pd
+import numpy as np
 import logging
 from collections import defaultdict
 
-import numpy as np
-import pandas as pd
-
 from dice_ml.data_interfaces.base_data_interface import _BaseData
-from dice_ml.utils.exception import (SystemException,
-                                     UserConfigValidationException)
+from dice_ml.utils.exception import SystemException, UserConfigValidationException
 
 
 class PublicData(_BaseData):
@@ -260,7 +258,7 @@ def get_valid_feature_range(self, feature_range_input, normalized=True):
         """
         feature_range = {}
 
-        for _, feature_name in enumerate(self.feature_names):
+        for idx, feature_name in enumerate(self.feature_names):
             feature_range[feature_name] = []
             if feature_name in self.continuous_feature_names:
                 max_value = self.data_df[feature_name].max()

dice_ml/dice.py

@@ -3,9 +3,9 @@
    such as RandomSampling, DiCEKD or DiCEGenetic"""
 
 from dice_ml.constants import BackEndTypes, SamplingStrategy
-from dice_ml.data_interfaces.private_data_interface import PrivateData
-from dice_ml.explainer_interfaces.explainer_base import ExplainerBase
 from dice_ml.utils.exception import UserConfigValidationException
+from dice_ml.explainer_interfaces.explainer_base import ExplainerBase
+from dice_ml.data_interfaces.private_data_interface import PrivateData
 
 
 class Dice(ExplainerBase):
@@ -67,14 +67,12 @@ def decide(model_interface, method):
 
     elif model_interface.backend == BackEndTypes.Tensorflow1:
         # pretrained Keras Sequential model with Tensorflow 1.x backend
-        from dice_ml.explainer_interfaces.dice_tensorflow1 import \
-            DiceTensorFlow1
+        from dice_ml.explainer_interfaces.dice_tensorflow1 import DiceTensorFlow1
         return DiceTensorFlow1
 
     elif model_interface.backend == BackEndTypes.Tensorflow2:
         # pretrained Keras Sequential model with Tensorflow 2.x backend
-        from dice_ml.explainer_interfaces.dice_tensorflow2 import \
-            DiceTensorFlow2
+        from dice_ml.explainer_interfaces.dice_tensorflow2 import DiceTensorFlow2
         return DiceTensorFlow2
 
     elif model_interface.backend == BackEndTypes.Pytorch:

dice_ml/diverse_counterfactuals.py

@@ -1,10 +1,8 @@
+import pandas as pd
 import copy
 import json
-
-import pandas as pd
-
-from dice_ml.constants import ModelTypes, _SchemaVersions
 from dice_ml.utils.serialize import DummyDataInterface
+from dice_ml.constants import _SchemaVersions, ModelTypes
 
 
 class _DiverseCFV1SchemaConstants:
@@ -117,7 +115,6 @@ def _visualize_internal(self, display_sparse_df=True, show_only_changes=False,
 
     def visualize_as_dataframe(self, display_sparse_df=True, show_only_changes=False):
         from IPython.display import display
-
         # original instance
         print('Query instance (original outcome : %i)' % round(self.test_pred))
         display(self.test_instance_df)  # works only in Jupyter notebook

dice_ml/explainer_interfaces/dice_KD.py

@@ -2,15 +2,14 @@
 Module to generate counterfactual explanations from a KD-Tree
 This code is similar to 'Interpretable Counterfactual Explanations Guided by Prototypes': https://arxiv.org/pdf/1907.02584.pdf
 """
-import copy
-import timeit
-
+from dice_ml.explainer_interfaces.explainer_base import ExplainerBase
 import numpy as np
+import timeit
 import pandas as pd
+import copy
 
 from dice_ml import diverse_counterfactuals as exp
 from dice_ml.constants import ModelTypes
-from dice_ml.explainer_interfaces.explainer_base import ExplainerBase
 
 
 class DiceKD(ExplainerBase):
@@ -260,10 +259,14 @@ def find_counterfactuals(self, data_df_copy, query_instance, query_instance_orig
             if total_cfs_found < total_CFs:
                 self.elapsed = timeit.default_timer() - start_time
                 m, s = divmod(self.elapsed, 60)
-                print('Only %d (required %d) ' % (total_cfs_found, self.total_CFs),
+                print('Only %d (required %d) ' % (total_cfs_found, total_CFs),
                       'Diverse Counterfactuals found for the given configuation, perhaps ',
                       'change the query instance or the features to vary...'  '; total time taken: %02d' % m,
                       'min %02d' % s, 'sec')
+            elif total_cfs_found == 0:
+                 print(
+                        'No Counterfactuals found for the given configuration, perhaps try with different parameters...',
+                        '; total time taken: %02d' % m, 'min %02d' % s, 'sec')
             else:
                 print('Diverse Counterfactuals found! total time taken: %02d' % m, 'min %02d' % s, 'sec')
 

dice_ml/explainer_interfaces/dice_genetic.py

@@ -2,17 +2,16 @@
 Module to generate diverse counterfactual explanations based on genetic algorithm
 This code is similar to 'GeCo: Quality Counterfactual Explanations in Real Time': https://arxiv.org/pdf/2101.01292.pdf
 """
-import copy
-import random
-import timeit
-
+from dice_ml.explainer_interfaces.explainer_base import ExplainerBase
 import numpy as np
 import pandas as pd
+import random
+import timeit
+import copy
 from sklearn.preprocessing import LabelEncoder
 
 from dice_ml import diverse_counterfactuals as exp
 from dice_ml.constants import ModelTypes
-from dice_ml.explainer_interfaces.explainer_base import ExplainerBase
 
 
 class DiceGenetic(ExplainerBase):
@@ -116,9 +115,8 @@ def do_random_init(self, num_inits, features_to_vary, query_instance, desired_cl
     def do_KD_init(self, features_to_vary, query_instance, cfs, desired_class, desired_range):
         cfs = self.label_encode(cfs)
         cfs = cfs.reset_index(drop=True)
-
-        self.cfs = np.zeros((self.population_size, self.data_interface.number_of_features))
-        for kx in range(self.population_size):
+        row = []
+        for kx in range(self.population_size*5):
             if kx >= len(cfs):
                 break
             one_init = np.zeros(self.data_interface.number_of_features)
@@ -143,16 +141,21 @@ def do_KD_init(self, features_to_vary, query_instance, cfs, desired_class, desir
                                 one_init[jx] = query_instance[jx]
                             else:
                                 one_init[jx] = np.random.choice(self.feature_range[feature])
-            self.cfs[kx] = one_init
+            t = tuple(one_init)
+            if t not in row:
+                row.append(t)
+                if len(row) == self.population_size:
+                    break
             kx += 1
+        self.cfs = np.array(row)
 
-        new_array = [tuple(row) for row in self.cfs]
-        uniques = np.unique(new_array, axis=0)
-
-        if len(uniques) != self.population_size:
+        #if len(self.cfs) > self.population_size:
+        #    pass
+        if len(self.cfs) != self.population_size:
+            print("Pericolo Loop infinito....!!!!")
             remaining_cfs = self.do_random_init(
-                self.population_size - len(uniques), features_to_vary, query_instance, desired_class, desired_range)
-            self.cfs = np.concatenate([uniques, remaining_cfs])
+                self.population_size - len(self.cfs), features_to_vary, query_instance, desired_class, desired_range)
+            self.cfs = np.concatenate([self.cfs, remaining_cfs])
 
     def do_cf_initializations(self, total_CFs, initialization, algorithm, features_to_vary, desired_range,
                               desired_class,
@@ -260,7 +263,7 @@ def _generate_counterfactuals(self, query_instance, total_CFs, initialization="k
                  (see diverse_counterfactuals.py).
         """
 
-        self.population_size = 10 * total_CFs
+        self.population_size = 3 * total_CFs
 
         self.start_time = timeit.default_timer()
 
@@ -514,6 +517,9 @@ def find_counterfactuals(self, query_instance, desired_range, desired_class,
             if len(self.final_cfs) == self.total_CFs:
                 print('Diverse Counterfactuals found! total time taken: %02d' %
                       m, 'min %02d' % s, 'sec')
+            elif len(self.final_cfs) == 0:
+                print('No Counterfactuals found for the given configuration, perhaps try with different parameters...',
+                      '; total time taken: %02d' % m, 'min %02d' % s, 'sec')
             else:
                 print('Only %d (required %d) ' % (len(self.final_cfs), self.total_CFs),
                       'Diverse Counterfactuals found for the given configuation, perhaps ',

dice_ml/explainer_interfaces/dice_pytorch.py

@@ -1,16 +1,16 @@
 """
 Module to generate diverse counterfactual explanations based on PyTorch framework
 """
-import copy
-import random
-import timeit
+from dice_ml.explainer_interfaces.explainer_base import ExplainerBase
+import torch
 
 import numpy as np
-import torch
+import random
+import timeit
+import copy
 
 from dice_ml import diverse_counterfactuals as exp
 from dice_ml.counterfactual_explanations import CounterfactualExplanations
-from dice_ml.explainer_interfaces.explainer_base import ExplainerBase
 
 
 class DicePyTorch(ExplainerBase):

dice_ml/explainer_interfaces/dice_random.py

@@ -3,15 +3,14 @@
 Module to generate diverse counterfactual explanations based on random sampling.
 A simple implementation.
 """
-import random
-import timeit
-
+from dice_ml.explainer_interfaces.explainer_base import ExplainerBase
 import numpy as np
 import pandas as pd
+import random
+import timeit
 
 from dice_ml import diverse_counterfactuals as exp
 from dice_ml.constants import ModelTypes
-from dice_ml.explainer_interfaces.explainer_base import ExplainerBase
 
 
 class DiceRandom(ExplainerBase):
@@ -109,11 +108,17 @@ class of query_instance for binary classification.
         cfs_df = None
         candidate_cfs = pd.DataFrame(
             np.repeat(query_instance.values, sample_size, axis=0), columns=query_instance.columns)
-        # Loop to change one feature at a time, then two features, and so on.
+        # Loop to change one feature at a time ##->(NOT TRUE), then two features, and so on.
         for num_features_to_vary in range(1, len(self.features_to_vary)+1):
+            # commented lines allow more values to change as num_features_to_vary increases, instead of .at you should use .loc
+            # is deliberately left commented out to let you choose.
+            # is slower, but more complete and still faster than genetic/KDtree
+            # selected_features = np.random.choice(self.features_to_vary, (sample_size, num_features_to_vary), replace=True)
             selected_features = np.random.choice(self.features_to_vary, (sample_size, 1), replace=True)
             for k in range(sample_size):
-                candidate_cfs.at[k, selected_features[k][0]] = random_instances.at[k, selected_features[k][0]]
+                candidate_cfs.at[k, selected_features[k][0]] = random_instances._get_value(k, selected_features[k][0])
+                # If you only want to change one feature, you should use _get_value
+                # candidate_cfs.iloc[k][selected_features[k]]=random_instances.iloc[k][selected_features[k]]
             scores = self.predict_fn(candidate_cfs)
             validity = self.decide_cf_validity(scores)
             if sum(validity) > 0: