Fix incompatible types in assignment

dice_ml/explainer_interfaces/dice_KD.py

@@ -163,7 +163,7 @@ def vary_valid(self, KD_query_instance, total_CFs, features_to_vary, permitted_r
 
         # TODO: this should be a user-specified parameter
         num_queries = min(len(self.dataset_with_predictions), total_CFs * 10)
-        cfs = []
+        cfs = pd.DataFrame()
 
         if self.KD_tree is not None and num_queries > 0:
             KD_tree_output = self.KD_tree.query(KD_query_instance, num_queries)

dice_ml/explainer_interfaces/dice_genetic.py

@@ -5,6 +5,7 @@
 import copy
 import random
 import timeit
+from typing import Any
 
 import numpy as np
 import pandas as pd
@@ -27,7 +28,7 @@ def __init__(self, data_interface, model_interface):
         self.num_output_nodes = None
 
         # variables required to generate CFs - see generate_counterfactuals() for more info
-        self.cfs = []
+        self.cfs = pd.DataFrame()
         self.features_to_vary = []
         self.cf_init_weights = []  # total_CFs, algorithm, features_to_vary
         self.loss_weights = []  # yloss_type, diversity_loss_type, feature_weights
@@ -343,12 +344,16 @@ def _predict_fn_custom(self, input_instance, desired_class):
 
     def compute_yloss(self, cfs, desired_range, desired_class):
         """Computes the first part (y-loss) of the loss function."""
-        yloss = 0.0
+        yloss: Any = 0.0
         if self.model.model_type == ModelTypes.Classifier:
             predicted_value = np.array(self.predict_fn_scores(cfs))
             if self.yloss_type == 'hinge_loss':
                 maxvalue = np.full((len(predicted_value)), -np.inf)
-                for c in range(self.num_output_nodes):
+                if self.num_output_nodes is None:
+                    num_output_nodes = 0
+                else:
+                    num_output_nodes = self.num_output_nodes
+                for c in range(num_output_nodes):
                     if c != desired_class:
                         maxvalue = np.maximum(maxvalue, predicted_value[:, c])
                 yloss = np.maximum(0, maxvalue - predicted_value[:, int(desired_class)])
@@ -429,7 +434,7 @@ def mate(self, k1, k2, features_to_vary, query_instance):
     def find_counterfactuals(self, query_instance, desired_range, desired_class,
                              features_to_vary, maxiterations, thresh, verbose):
         """Finds counterfactuals by generating cfs through the genetic algorithm"""
-        population = self.cfs.copy()
+        population: Any = self.cfs.copy()
         iterations = 0
         previous_best_loss = -np.inf
         current_best_loss = np.inf

dice_ml/explainer_interfaces/dice_pytorch.py

@@ -4,6 +4,7 @@
 import copy
 import random
 import timeit
+from typing import Any, Optional, Union
 
 import numpy as np
 import torch
@@ -223,14 +224,16 @@ def do_optimizer_initializations(self, optimizer, learning_rate):
         opt_method = optimizer.split(':')[1]
 
         # optimizater initialization
+        self.optimizer: Optional[Union[torch.optim.Adam, torch.optim.RMSprop]] = None
         if opt_method == "adam":
             self.optimizer = torch.optim.Adam(self.cfs, lr=learning_rate)
         elif opt_method == "rmsprop":
             self.optimizer = torch.optim.RMSprop(self.cfs, lr=learning_rate)
 
     def compute_yloss(self):
         """Computes the first part (y-loss) of the loss function."""
-        yloss = 0.0
+        yloss: Any = 0.0
+        criterion: Optional[Union[torch.nn.BCEWithLogitsLoss, torch.nn.ReLU]] = None
         for i in range(self.total_CFs):
             if self.yloss_type == "l2_loss":
                 temp_loss = torch.pow((self.get_model_output(self.cfs[i]) - self.target_cf_class), 2)[0]
@@ -307,7 +310,7 @@ def compute_diversity_loss(self):
     def compute_regularization_loss(self):
         """Adds a linear equality constraints to the loss functions -
            to ensure all levels of a categorical variable sums to one"""
-        regularization_loss = 0.0
+        regularization_loss: Any = 0.0
         for i in range(self.total_CFs):
             for v in self.encoded_categorical_feature_indexes:
                 regularization_loss += torch.pow((torch.sum(self.cfs[i][v[0]:v[-1]+1]) - 1.0), 2)
@@ -425,7 +428,7 @@ def find_counterfactuals(self, query_instance, desired_class, optimizer, learnin
         test_pred = self.predict_fn(torch.tensor(query_instance).float())[0]
         if desired_class == "opposite":
             desired_class = 1.0 - np.round(test_pred)
-        self.target_cf_class = torch.tensor(desired_class).float()
+        self.target_cf_class: Any = torch.tensor(desired_class).float()
 
         self.min_iter = min_iter
         self.max_iter = max_iter

dice_ml/explainer_interfaces/dice_random.py

@@ -30,10 +30,9 @@ def __init__(self, data_interface, model_interface):
         self.model.transformer.initialize_transform_func()
 
         self.precisions = self.data_interface.get_decimal_precisions(output_type="dict")
-        if self.data_interface.outcome_name in self.precisions:
-            self.outcome_precision = [self.precisions[self.data_interface.outcome_name]]
-        else:
-            self.outcome_precision = 0
+        self.outcome_precision = [
+            self.precisions[self.data_interface.outcome_name]
+        ] if self.data_interface.outcome_name in self.precisions else 0
 
     def _generate_counterfactuals(self, query_instance, total_CFs, desired_range=None,
                                   desired_class="opposite", permitted_range=None,

dice_ml/explainer_interfaces/dice_tensorflow2.py

@@ -341,8 +341,7 @@ def initialize_CFs(self, query_instance, init_near_query_instance=False):
                         one_init.append(np.random.uniform(self.minx[0][i], self.maxx[0][i]))
                 else:
                     one_init.append(query_instance[0][i])
-            one_init = np.array([one_init], dtype=np.float32)
-            self.cfs[n].assign(one_init)
+            self.cfs[n].assign(np.array([one_init], dtype=np.float32))
 
     def round_off_cfs(self, assign=False):
         """function for intermediate projection of CFs."""

dice_ml/explainer_interfaces/explainer_base.py

@@ -5,6 +5,7 @@
 import pickle
 from abc import ABC, abstractmethod
 from collections.abc import Iterable
+from typing import Any, Dict, Optional, Union
 
 import numpy as np
 import pandas as pd
@@ -152,10 +153,9 @@ def generate_counterfactuals(self, query_instances, total_CFs,
         cf_examples_arr = []
         query_instances_list = []
         if isinstance(query_instances, pd.DataFrame):
-            for ix in range(query_instances.shape[0]):
-                query_instances_list.append(query_instances[ix:(ix+1)])
+            query_instances_list = [query_instances[ix:(ix+1)] for ix in range(query_instances.shape[0])]
         elif isinstance(query_instances, Iterable):
-            query_instances_list = query_instances
+            query_instances_list = [query_instance for query_instance in query_instances]
         for query_instance in tqdm(query_instances_list):
             self.data_interface.set_continuous_feature_indexes(query_instance)
             res = self._generate_counterfactuals(
@@ -416,7 +416,7 @@ def feature_importance(self, query_instances, cf_examples_list=None,
                 posthoc_sparsity_algorithm=posthoc_sparsity_algorithm,
                 **kwargs).cf_examples_list
         allcols = self.data_interface.categorical_feature_names + self.data_interface.continuous_feature_names
-        summary_importance = None
+        summary_importance: Optional[Union[Dict[int, float]]] = None
         local_importances = None
         if global_importance:
             summary_importance = {}
@@ -731,13 +731,16 @@ def is_cf_valid(self, model_score):
             model_score = model_score[0]
         # Converting target_cf_class to a scalar (tf/torch have it as (1,1) shape)
         if self.model.model_type == ModelTypes.Classifier:
-            target_cf_class = self.target_cf_class
             if hasattr(self.target_cf_class, "shape"):
                 if len(self.target_cf_class.shape) == 1:
-                    target_cf_class = self.target_cf_class[0]
+                    temp_target_cf_class = self.target_cf_class[0]
                 elif len(self.target_cf_class.shape) == 2:
-                    target_cf_class = self.target_cf_class[0][0]
-            target_cf_class = int(target_cf_class)
+                    temp_target_cf_class = self.target_cf_class[0][0]
+                else:
+                    temp_target_cf_class = int(self.target_cf_class)
+            else:
+                temp_target_cf_class = int(self.target_cf_class)
+            target_cf_class = temp_target_cf_class
 
             if len(model_score) == 1:  # for tensorflow/pytorch models
                 pred_1 = model_score[0]
@@ -757,6 +760,7 @@ def is_cf_valid(self, model_score):
             return self.target_cf_range[0] <= model_score and model_score <= self.target_cf_range[1]
 
     def get_model_output_from_scores(self, model_scores):
+        output_type: Any = None
         if self.model.model_type == ModelTypes.Classifier:
             output_type = np.int32
         else:
@@ -806,17 +810,19 @@ def build_KD_tree(self, data_df_copy, desired_range, desired_class, predicted_ou
         data_df_copy[predicted_outcome_name] = predictions
 
         # segmenting the dataset according to outcome
-        dataset_with_predictions = None
         if self.model.model_type == ModelTypes.Classifier:
             dataset_with_predictions = data_df_copy.loc[[i == desired_class for i in predictions]].copy()
 
         elif self.model.model_type == ModelTypes.Regressor:
             dataset_with_predictions = data_df_copy.loc[
                 [desired_range[0] <= pred <= desired_range[1] for pred in predictions]].copy()
 
+        else:
+            dataset_with_predictions = None
+
         KD_tree = None
         # Prepares the KD trees for DiCE
-        if len(dataset_with_predictions) > 0:
+        if dataset_with_predictions is not None and len(dataset_with_predictions) > 0:
             dummies = pd.get_dummies(dataset_with_predictions[self.data_interface.feature_names])
             KD_tree = KDTree(dummies)
 

dice_ml/explainer_interfaces/feasible_base_vae.py

@@ -136,8 +136,9 @@ def train(self, pre_trained=False):
             train_loss = 0.0
             train_size = 0
 
-            train_dataset = torch.tensor(self.vae_train_feat).float()
-            train_dataset = torch.utils.data.DataLoader(train_dataset, batch_size=self.batch_size, shuffle=True)
+            train_dataset = torch.utils.data.DataLoader(
+                torch.tensor(self.vae_train_feat).float(),  # type: ignore
+                batch_size=self.batch_size, shuffle=True)
             for train in enumerate(train_dataset):
                 self.cf_vae_optimizer.zero_grad()
 
@@ -178,8 +179,7 @@ def generate_counterfactuals(self, query_instance, total_CFs, desired_class="opp
         final_cf_pred = []
         final_test_pred = []
         for i in range(len(query_instance)):
-            train_x = test_dataset[i]
-            train_x = torch.tensor(train_x).float()
+            train_x = torch.tensor(test_dataset[i]).float()
             train_y = torch.argmax(self.pred_model(train_x), dim=1)
 
             curr_gen_cf = []

dice_ml/explainer_interfaces/feasible_model_approx.py

@@ -81,8 +81,9 @@ def train(self, constraint_type, constraint_variables, constraint_direction, con
             train_loss = 0.0
             train_size = 0
 
-            train_dataset = torch.tensor(self.vae_train_feat).float()
-            train_dataset = torch.utils.data.DataLoader(train_dataset, batch_size=self.batch_size, shuffle=True)
+            train_dataset = torch.utils.data.DataLoader(
+                torch.tensor(self.vae_train_feat).float(),  # type: ignore
+                batch_size=self.batch_size, shuffle=True)
             for train in enumerate(train_dataset):
                 self.cf_vae_optimizer.zero_grad()