Source code for dice_ml.data_interfaces.public_data_interface

"""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
from sklearn.model_selection import train_test_split
import logging
from collections import defaultdict


[docs]class PublicData: """A data interface for public data. This class is an interface to DiCE explainers and contains methods to transform user-fed raw data into the format a DiCE explainer requires, and vice versa.""" def __init__(self, params): """Init method :param dataframe: The train dataframe used by explainer method. :param continuous_features: List of names of continuous features. The remaining features are categorical features. :param outcome_name: Outcome feature name. :param permitted_range (optional): Dictionary with feature names as keys and permitted range in list as values. Defaults to the range inferred from training data. :param continuous_features_precision (optional): Dictionary with feature names as keys and precisions as values. :param data_name (optional): Dataset name """ if isinstance(params['dataframe'], pd.DataFrame): self.data_df = params['dataframe'] else: raise ValueError("should provide a pandas dataframe") if type(params['continuous_features']) is list: self.continuous_feature_names = params['continuous_features'] else: raise ValueError( "should provide the name(s) of continuous features in the data") if type(params['outcome_name']) is str: self.outcome_name = params['outcome_name'] else: raise ValueError("should provide the name of outcome feature") self.categorical_feature_names = [name for name in self.data_df.columns.tolist( ) if name not in self.continuous_feature_names + [self.outcome_name]] self.feature_names = [ name for name in self.data_df.columns.tolist() if name != self.outcome_name] self.continuous_feature_indexes = [self.data_df.columns.get_loc( name) for name in self.continuous_feature_names if name in self.data_df] self.categorical_feature_indexes = [self.data_df.columns.get_loc( name) for name in self.categorical_feature_names if name in self.data_df] if 'continuous_features_precision' in params: self.continuous_features_precision = params['continuous_features_precision'] else: self.continuous_features_precision = None if len(self.categorical_feature_names) > 0: for feature in self.categorical_feature_names: self.data_df[feature] = self.data_df[feature].apply(str) self.data_df[self.categorical_feature_names] = self.data_df[self.categorical_feature_names].astype( 'category') if len(self.continuous_feature_names) > 0: for feature in self.continuous_feature_names: if self.get_data_type(feature) == 'float': self.data_df[feature] = self.data_df[feature].astype( np.float32) else: self.data_df[feature] = self.data_df[feature].astype( np.int32) # should move the below snippet to gradient based dice interfaces # self.one_hot_encoded_data = self.one_hot_encode_data(self.data_df) # self.ohe_encoded_feature_names = [x for x in self.one_hot_encoded_data.columns.tolist( # ) if x not in np.array([self.outcome_name])] # should move the below snippet to model agnostic dice interfaces # # Initializing a label encoder to obtain label-encoded values for categorical variables # self.labelencoder = {} # # self.label_encoded_data = self.data_df.copy() # # for column in self.categorical_feature_names: # self.labelencoder[column] = LabelEncoder() # self.label_encoded_data[column] = self.labelencoder[column].fit_transform(self.data_df[column]) input_permitted_range = None if 'permitted_range' in params: input_permitted_range = params['permitted_range'] self.permitted_range, feature_ranges_orig = self.get_features_range(input_permitted_range) # should move the below snippet to model agnostic dice interfaces # self.max_range = -np.inf # for feature in self.continuous_feature_names: # self.max_range = max(self.max_range, self.permitted_range[feature][1]) if 'data_name' in params: self.data_name = params['data_name'] else: self.data_name = 'mydata'
[docs] def get_features_range(self, permitted_range_input=None): ranges = {} # Getting default ranges based on the dataset for feature_name in self.continuous_feature_names: ranges[feature_name] = [ self.data_df[feature_name].min(), self.data_df[feature_name].max()] for feature_name in self.categorical_feature_names: ranges[feature_name] = self.data_df[feature_name].unique().tolist() feature_ranges_orig = ranges.copy() # Overwriting the ranges for a feature if input provided if permitted_range_input is not None: for feature_name, feature_range in permitted_range_input.items(): ranges[feature_name] = feature_range return ranges, feature_ranges_orig
[docs] def get_data_type(self, col): """Infers data type of a continuous feature from the training data.""" if ((self.data_df[col].dtype == np.int64) or (self.data_df[col].dtype == np.int32)): return 'int' elif ((self.data_df[col].dtype == np.float64) or (self.data_df[col].dtype == np.float32)): return 'float' else: raise ValueError("Unknown data type of feature %s: must be int or float" % col)
[docs] def one_hot_encode_data(self, data): """One-hot-encodes the data.""" return pd.get_dummies(data, drop_first=False, columns=self.categorical_feature_names)
[docs] def normalize_data(self, df): """Normalizes continuous features to make them fall in the range [0,1].""" result = df.copy() for feature_name in self.continuous_feature_names: max_value = self.data_df[feature_name].max() min_value = self.data_df[feature_name].min() result[feature_name] = ( df[feature_name] - min_value) / (max_value - min_value) #if encoding == 'label': # for ix in self.categorical_feature_indexes: # feature_name = self.feature_names[ix] # max_value = len(self.train_df[feature_name].unique())-1 # min_value = 0 # result[feature_name] = (df[feature_name] - min_value) / (max_value - min_value) return result
[docs] def de_normalize_data(self, df): """De-normalizes continuous features from [0,1] range to original range.""" if len(df) == 0: return df result = df.copy() for feature_name in self.continuous_feature_names: max_value = self.data_df[feature_name].max() min_value = self.data_df[feature_name].min() result[feature_name] = ( df[feature_name] * (max_value - min_value)) + min_value return result
[docs] def get_valid_feature_range(self, feature_range_input, normalized=True): """Gets the min/max value of features in normalized or de-normalized form. Assumes that all features are already encoded to numerical form such that the number of features remains the same. # TODO needs work adhere to label encoded max and to support permitted_range for both continuous and discrete when provided in _generate_counterfactuals. """ feature_range = {} 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() min_value = self.data_df[feature_name].min() if normalized: minx = (feature_range_input[feature_name] [0] - min_value) / (max_value - min_value) maxx = (feature_range_input[feature_name] [1] - min_value) / (max_value - min_value) else: minx = feature_range_input[feature_name][0] maxx = feature_range_input[feature_name][1] feature_range[feature_name].append(minx) feature_range[feature_name].append(maxx) else: # categorical features feature_range[feature_name] = feature_range_input[feature_name] return feature_range
[docs] def get_minx_maxx(self, normalized=True): """Gets the min/max value of features in normalized or de-normalized form.""" minx = np.array([[0.0] * len(self.ohe_encoded_feature_names)]) maxx = np.array([[1.0] * len(self.ohe_encoded_feature_names)]) for idx, feature_name in enumerate(self.continuous_feature_names): max_value = self.data_df[feature_name].max() min_value = self.data_df[feature_name].min() if normalized: minx[0][idx] = (self.permitted_range[feature_name] [0] - min_value) / (max_value - min_value) maxx[0][idx] = (self.permitted_range[feature_name] [1] - min_value) / (max_value - min_value) else: minx[0][idx] = self.permitted_range[feature_name][0] maxx[0][idx] = self.permitted_range[feature_name][1] return minx, maxx
#if encoding=='one-hot': # minx = np.array([[0.0] * len(self.ohe_encoded_feature_names)]) # maxx = np.array([[1.0] * len(self.ohe_encoded_feature_names)]) # for idx, feature_name in enumerate(self.continuous_feature_names): # max_value = self.train_df[feature_name].max() # min_value = self.train_df[feature_name].min() # if normalized: # minx[0][idx] = (self.permitted_range[feature_name] # [0] - min_value) / (max_value - min_value) # maxx[0][idx] = (self.permitted_range[feature_name] # [1] - min_value) / (max_value - min_value) # else: # minx[0][idx] = self.permitted_range[feature_name][0] # maxx[0][idx] = self.permitted_range[feature_name][1] #else: # minx = np.array([[0.0] * len(self.feature_names)]) # maxx = np.array([[1.0] * len(self.feature_names)])
[docs] def get_mads(self, normalized=False): """Computes Median Absolute Deviation of features.""" mads = {} if normalized is False: for feature in self.continuous_feature_names: mads[feature] = np.median( abs(self.data_df[feature].values - np.median(self.data_df[feature].values))) else: normalized_train_df = self.normalize_data(self.data_df) for feature in self.continuous_feature_names: mads[feature] = np.median( abs(normalized_train_df[feature].values - np.median(normalized_train_df[feature].values))) return mads
[docs] def get_valid_mads(self, normalized=False, display_warnings=False, return_mads=True): """Computes Median Absolute Deviation of features. If they are <=0, returns a practical value instead""" mads = self.get_mads(normalized=normalized) for feature in mads: if mads[feature] <= 0: mads[feature] = 1.0 if display_warnings: logging.warning(" MAD for feature %s is 0, so replacing it with 1.0 to avoid error.", feature) if return_mads: return mads
[docs] def get_quantiles_from_training_data(self, quantile=0.05, normalized=False): """Computes required quantile of Absolute Deviations of features.""" quantiles = {} if normalized is False: for feature in self.continuous_feature_names: quantiles[feature] = np.quantile( abs(list(set(self.data_df[feature].tolist())) - np.median( list(set(self.data_df[feature].tolist())))), quantile) else: normalized_train_df = self.normalize_data(self.data_df) for feature in self.continuous_feature_names: quantiles[feature] = np.quantile( abs(list(set(normalized_train_df[feature].tolist())) - np.median( list(set(normalized_train_df[feature].tolist())))), quantile) return quantiles
[docs] def create_ohe_params(self): if len(self.categorical_feature_names) > 0: one_hot_encoded_data = self.one_hot_encode_data(self.data_df) self.ohe_encoded_feature_names = [x for x in one_hot_encoded_data.columns.tolist( ) if x not in np.array([self.outcome_name])] else: # one-hot-encoded data is same as original data if there is no categorical features. self.ohe_encoded_feature_names = [feat for feat in self.feature_names] self.ohe_base_df = self.prepare_df_for_ohe_encoding() # base dataframe for doing one-hot-encoding
# ohe_encoded_feature_names and ohe_base_df are created (and stored as data class's parameters) when get_data_params_for_gradient_dice() is called from gradient-based DiCE explainers
[docs] def get_data_params_for_gradient_dice(self): """Gets all data related params for DiCE.""" self.create_ohe_params() minx, maxx = self.get_minx_maxx(normalized=True) # get the column indexes of categorical and continuous features after one-hot-encoding encoded_categorical_feature_indexes = self.get_encoded_categorical_feature_indexes() flattened_indexes = [item for sublist in encoded_categorical_feature_indexes for item in sublist] encoded_continuous_feature_indexes = [ix for ix in range(len(minx[0])) if ix not in flattened_indexes] # min and max for continuous features in original scale org_minx, org_maxx = self.get_minx_maxx(normalized=False) cont_minx = list(org_minx[0][encoded_continuous_feature_indexes]) cont_maxx = list(org_maxx[0][encoded_continuous_feature_indexes]) # decimal precisions for continuous features cont_precisions = [self.get_decimal_precisions()[ix] for ix in range(len(self.continuous_feature_names))] return minx, maxx, encoded_categorical_feature_indexes, encoded_continuous_feature_indexes, cont_minx, cont_maxx, cont_precisions
[docs] def get_encoded_categorical_feature_indexes(self): """Gets the column indexes categorical features after one-hot-encoding.""" cols = [] for col_parent in self.categorical_feature_names: temp = [self.ohe_encoded_feature_names.index( col) for col in self.ohe_encoded_feature_names if col.startswith(col_parent) and col not in self.continuous_feature_names] cols.append(temp) return cols
[docs] def get_indexes_of_features_to_vary(self, features_to_vary='all'): """Gets indexes from feature names of one-hot-encoded data.""" # TODO: add encoding as a parameter and use the function get_indexes_of_features_to_vary for label encoding too if features_to_vary == "all": return [i for i in range(len(self.ohe_encoded_feature_names))] else: ixs = [] encoded_cats_ixs = self.get_encoded_categorical_feature_indexes() encoded_cats_ixs = [item for sublist in encoded_cats_ixs for item in sublist] for colidx, col in enumerate(self.ohe_encoded_feature_names): if colidx in encoded_cats_ixs and col.startswith(tuple(features_to_vary)): ixs.append(colidx) elif colidx not in encoded_cats_ixs and col in features_to_vary: ixs.append(colidx) return ixs
[docs] def from_label(self, data): """Transforms label encoded data back to categorical values""" out = data.copy() if isinstance(data, pd.DataFrame) or isinstance(data, dict): for column in self.categorical_feature_names: out[column] = self.labelencoder[column].inverse_transform(out[column].round().astype(int).tolist()) return out elif isinstance(data, list): for c in self.categorical_feature_indexes: out[c] = self.labelencoder[self.feature_names[c]].inverse_transform([round(out[c])])[0] return out
[docs] def from_dummies(self, data, prefix_sep='_'): """Gets the original data from dummy encoded data with k levels.""" out = data.copy() for feat in self.categorical_feature_names: # first, derive column names in the one-hot-encoded data from the original data cat_col_values = [] for val in list(self.data_df[feat].unique()): cat_col_values.append(feat + prefix_sep + str( val)) # join original feature name and its unique values , ex: education_school match_cols = [c for c in data.columns if c in cat_col_values] # check for the above matching columns in the encoded data # then, recreate original data by removing the suffixes - based on the GitHub issue comment: https://github.com/pandas-dev/pandas/issues/8745#issuecomment-417861271 cols, labs = [[c.replace( x, "") for c in match_cols] for x in ["", feat + prefix_sep]] out[feat] = pd.Categorical( np.array(labs)[np.argmax(data[cols].values, axis=1)]) out.drop(cols, axis=1, inplace=True) return out
[docs] def get_decimal_precisions(self, output_type="list"): """"Gets the precision of continuous features in the data.""" # if the precision of a continuous feature is not given, we use the maximum precision of the modes to capture the precision of majority of values in the column. precisions_dict = defaultdict(int) precisions = [0] * len(self.feature_names) for ix, col in enumerate(self.continuous_feature_names): if ((self.continuous_features_precision is not None) and (col in self.continuous_features_precision)): precisions[ix] = self.continuous_features_precision[col] precisions_dict[col] = self.continuous_features_precision[col] elif ((self.data_df[col].dtype == np.float32) or (self.data_df[col].dtype == np.float64)): modes = self.data_df[col].mode() maxp = len(str(modes[0]).split('.')[1]) # maxp stores the maximum precision of the modes for mx in range(len(modes)): prec = len(str(modes[mx]).split('.')[1]) if prec > maxp: maxp = prec precisions[ix] = maxp precisions_dict[col] = maxp if output_type == "list": return precisions elif output_type == "dict": return precisions_dict
[docs] def get_decoded_data(self, data, encoding='one-hot'): """Gets the original data from encoded data.""" if len(data) == 0: return data index = [i for i in range(0, len(data))] if encoding == 'one-hot': if isinstance(data, pd.DataFrame): return self.from_dummies(data) elif isinstance(data, np.ndarray): data = pd.DataFrame(data=data, index=index, columns=self.ohe_encoded_feature_names) return self.from_dummies(data) else: raise ValueError("data should be a pandas dataframe or a numpy array") elif encoding == 'label': data = pd.DataFrame(data=data, index=index, columns=self.feature_names) return data
[docs] def prepare_df_for_ohe_encoding(self): """Create base dataframe to do OHE for a single instance or a set of instances""" levels = [] colnames = [feat for feat in self.categorical_feature_names] for cat_feature in colnames: levels.append(self.data_df[cat_feature].cat.categories.tolist()) if len(colnames) > 0: df = pd.DataFrame({colnames[0]: levels[0]}) else: df = pd.DataFrame() for col in range(1, len(colnames)): temp_df = pd.DataFrame({colnames[col]: levels[col]}) df = pd.concat([df, temp_df], axis=1, sort=False) colnames = [feat for feat in self.continuous_feature_names] for col in range(0, len(colnames)): temp_df = pd.DataFrame({colnames[col]: []}) df = pd.concat([df, temp_df], axis=1, sort=False) return df
[docs] def prepare_query_instance(self, query_instance): """Prepares user defined test input(s) for DiCE.""" if isinstance(query_instance, list): if isinstance(query_instance[0], dict): # prepare a list of query instances test = pd.DataFrame(query_instance, columns=self.feature_names) else: # prepare a single query instance in list query_instance = {'row1': query_instance} test = pd.DataFrame.from_dict( query_instance, orient='index', columns=self.feature_names) elif isinstance(query_instance, dict): test = pd.DataFrame({k: [v] for k, v in query_instance.items()}, columns=self.feature_names) elif isinstance(query_instance, pd.DataFrame): test = query_instance.copy() else: raise ValueError("Query instance should be a dict, a pandas dataframe, a list, or a list of dicts") test = test.reset_index(drop=True) return test
# TODO: create a new method, get_LE_min_max_normalized_data() to get label-encoded and normalized data. Keep this method only for converting query_instance to pd.DataFrame # if encoding == 'label': # for column in self.categorical_feature_names: # test[column] = self.labelencoder[column].transform(test[column]) # return self.normalize_data(test, encoding) # # elif encoding == 'one-hot': # temp = self.prepare_df_for_encoding() # temp = temp.append(test, ignore_index=True, sort=False) # temp = self.one_hot_encode_data(temp) # temp = self.normalize_data(temp) # # return temp.tail(test.shape[0]).reset_index(drop=True)
[docs] def get_ohe_min_max_normalized_data(self, query_instance): """Transforms query_instance into one-hot-encoded and min-max normalized data. query_instance should be a dict, a dataframe, a list, or a list of dicts""" query_instance = self.prepare_query_instance(query_instance) temp = self.ohe_base_df.append(query_instance, ignore_index=True, sort=False) temp = self.one_hot_encode_data(temp) temp = temp.tail(query_instance.shape[0]).reset_index(drop=True) return self.normalize_data(temp) # returns a pandas dataframe
[docs] def get_inverse_ohe_min_max_normalized_data(self, transformed_data): """Transforms one-hot-encoded and min-max normalized data into raw user-fed data format. transformed_data should be a dataframe or an array""" raw_data = self.get_decoded_data(transformed_data, encoding='one-hot') raw_data = self.de_normalize_data(raw_data) precisions = self.get_decimal_precisions() for ix, feature in enumerate(self.continuous_feature_names): raw_data[feature] = raw_data[feature].astype(float).round(precisions[ix]) raw_data = raw_data[self.feature_names] return raw_data # returns a pandas dataframe