def _calculate_p_value(feature_column, y, settings, target_is_binary): """ Internal helper function to calculate the p-value of a given feature using one of the dedicated functions target_*_feature_*_test. :param feature_column: the feature column. :type feature_column: pandas.Series :param y: the binary target vector :type y: pandas.Series :param settings: The settings object to control how the significance is calculated. :type settings: FeatureSignificanceTestsSettings :param target_is_binary: Whether the target is binary or not :type target_is_binary: bool :return: the p-value of the feature significance test and the type of the tested feature as a Series. Lower p-values indicate a higher feature significance. :rtype: pd.Series """ # Do not process constant features if len(pd.unique(feature_column.values)) == 1: _logger.warning( "[test_feature_significance] Feature {} is constant".format( feature_column.name)) return pd.Series({ "type": "const", "rejected": False }, name=feature_column.name) else: if target_is_binary: # Decide if the current feature is binary or not if len(set(feature_column.values)) == 2: type = "binary" p_value = target_binary_feature_binary_test( feature_column, y, settings) else: type = "real" p_value = target_binary_feature_real_test( feature_column, y, settings) else: # Decide if the current feature is binary or not if len(set(feature_column.values)) == 2: type = "binary" p_value = target_real_feature_binary_test( feature_column, y, settings) else: type = "real" p_value = target_real_feature_real_test( feature_column, y, settings) return pd.Series({ "p_value": p_value, "type": type }, name=feature_column.name)
def _calculate_p_value(feature_column, y, target_is_binary, test_for_binary_target_real_feature): """ Internal helper function to calculate the p-value of a given feature using one of the dedicated functions target_*_feature_*_test. :param feature_column: the feature column. :type feature_column: pandas.Series :param y: the binary target vector :type y: pandas.Series :param target_is_binary: Whether the target is binary or not :type target_is_binary: bool :param test_for_binary_target_real_feature: The significance test to be used for binary target and real valued features. Either ``'mann'`` for the Mann-Whitney-U test or ``'smir'`` for the Kolmogorov-Smirnov test. :type test_for_binary_target_real_feature: str :return: the p-value of the feature significance test and the type of the tested feature as a Series. Lower p-values indicate a higher feature significance. :rtype: pd.Series """ # Do not process constant features if len(pd.unique(feature_column.values)) == 1: _logger.warning( "[test_feature_significance] Feature {} is constant".format( feature_column.name)) return pd.Series({ "type": "const", "rejected": False }, name=feature_column.name) else: if target_is_binary: # Decide if the current feature is binary or not if len(set(feature_column.values)) == 2: type = "binary" p_value = target_binary_feature_binary_test(feature_column, y) else: type = "real" p_value = target_binary_feature_real_test( feature_column, y, test_for_binary_target_real_feature) else: # Decide if the current feature is binary or not if len(set(feature_column.values)) == 2: type = "binary" p_value = target_real_feature_binary_test(feature_column, y) else: type = "real" p_value = target_real_feature_real_test(feature_column, y) return pd.Series({ "p_value": p_value, "type": type }, name=feature_column.name)
def test_feature_selection_target_realvalued_features_realvalued(self, minimal_p_value_for_unsignificant_features, real_feature, real_target_not_related): """ Test if the p_value returned by target_real_feature_real_test is large enough for highly unsignificant features. """ p_value = target_real_feature_real_test(real_feature, real_target_not_related) assert minimal_p_value_for_unsignificant_features < p_value
def test_feature_selection_target_realvalued_features_realvalued(self, maximal_p_value_for_significant_features, real_feature): """ Test if the p_value returned by target_real_feature_real_test is low enough for highly significant features. """ y = real_feature + pd.Series(np.random.normal(0, 1, 250)) p_value = target_real_feature_real_test(real_feature, y) assert maximal_p_value_for_significant_features > p_value
def extract_time_features_DataFrame(Features_DataFrame, timeseries): # timeseries : according to the window size , extracted y values # Features_DataFrame : corresponding last values. # these two should have same time index time_features = pd.DataFrame(index=Features_DataFrame.index) time_features["nanosecond"] = Features_DataFrame.index.nanosecond time_features["second"] = Features_DataFrame.index.second time_features["minute"] = Features_DataFrame.index.minute time_features["hour"] = Features_DataFrame.index.hour time_features["week"] = Features_DataFrame.index.week time_features["weekday"] = Features_DataFrame.index.weekday time_features["month"] = Features_DataFrame.index.month time_features["year"] = Features_DataFrame.index.year time_features["day"] = Features_DataFrame.index.day scaler_index = [] # for year, nanosecond # drop the unique features time_features = time_features.loc[:, time_features.apply(pd.Series.nunique ) != 1] # drop irrelevant features columns = time_features.columns for column in columns: p = target_real_feature_real_test(time_features[column], timeseries) if p > 0.05: del time_features[column] for column in time_features.columns: if column not in scaled_index: period = check_date_number[column] / (2 * np.pi) time_features[column] = time_features[column] / period time_features['{}_sin'.format(column)] = np.around( (np.sin(time_features[column]) + 1) / 2, decimals=3) time_features['{}_cos'.format(column)] = np.around( (np.cos(time_features[column]) + 1) / 2, decimals=3) del time_features[column] scalers = OrderedDict() for column in time_features.columns: if column in scaled_index: array = np.array(time_features[column]) scaler = MinMaxScaler(feature_range=(0, 1)) scaled = scaler.fit_transform(array.reshape(-1, 1)).reshape(-1) time_features[column] = scaled scalers[column] = scaler return scalers, time_features, time_features.columns
def identify_and_remove_unique_columns(Dataframe): Dataframe = Dataframe.copy() del Dataframe["rtf_id"] del Dataframe["cycle"] unique_counts = Dataframe.nunique() record_single_unique = pd.DataFrame( unique_counts[unique_counts == 1]).reset_index().rename(columns={ 'index': 'feature', 0: 'nunique' }) unique_to_drop = list(record_single_unique['feature']) Dataframe = Dataframe.drop(columns=unique_to_drop) unique_counts = Dataframe.nunique() record_single_unique = pd.DataFrame(unique_counts).reset_index().rename( columns={ 'index': 'feature', 0: 'nunique' }) record_single_unique["type"] = record_single_unique["nunique"].apply( lambda x: "real" if x > 2 else "binary") for i in range(record_single_unique.shape[0]): col = record_single_unique.loc[i, "feature"] _type = record_single_unique.loc[i, "type"] if _type == "real": p_value = target_real_feature_real_test(Dataframe[col], Dataframe["RUL_pw"]) else: le = preprocessing.LabelEncoder() p_value = target_real_feature_binary_test( pd.Series(le.fit_transform(Dataframe[col])), Dataframe["RUL_pw"]) if p_value > 0.05: unique_to_drop.append(col) return unique_to_drop
def test_fs_tr_fr_series(self): with pytest.raises(TypeError): target_real_feature_real_test(x=[0, 1, 2], y=pd.Series([0, 1, 2])) with pytest.raises(TypeError): target_real_feature_real_test(x=pd.Series([0, 1, 2]), y=[0, 1, 2])
def test_checks_target_nan(self, real_series_with_nan, real_series): with pytest.raises(ValueError): target_real_feature_real_test(x=real_series, y=real_series_with_nan)
def test_checks_target_is_series(self, real_series): with pytest.raises(TypeError): target_real_feature_real_test(x=real_series, y=real_series.values)
def check_fs_sig_bh(X, y, settings=None): """ The wrapper function that calls the significance test functions in this package. In total, for each feature from the input pandas.DataFrame an univariate feature significance test is conducted. Those tests generate p values that are then evaluated by the Benjamini Hochberg procedure to decide which features to keep and which to delete. We are testing :math:`H_0` = the Feature is not relevant and can not be added against :math:`H_1` = the Feature is relevant and should be kept or in other words :math:`H_0` = Target and Feature are independent / the Feature has no influence on the target :math:`H_1` = Target and Feature are associated / dependent When the target is binary this becomes :math:`H_0 = \\left( F_{\\text{target}=1} = F_{\\text{target}=0} \\right)` :math:`H_1 = \\left( F_{\\text{target}=1} \\neq F_{\\text{target}=0} \\right)` Where :math:`F` is the distribution of the target. In the same way we can state the hypothesis when the feature is binary :math:`H_0 = \\left( T_{\\text{feature}=1} = T_{\\text{feature}=0} \\right)` :math:`H_1 = \\left( T_{\\text{feature}=1} \\neq T_{\\text{feature}=0} \\right)` Here :math:`T` is the distribution of the target. TODO: And for real valued? :param X: The DataFrame containing all the features and the target :type X: pandas.DataFrame :param y: The target vector :type y: pandas.Series :param settings: The feature selection settings to use for performing the tests. :type settings: FeatureSignificanceTestsSettings :return: A pandas.DataFrame with each column of the input DataFrame X as index with information on the significance of this particular feature. The DataFrame has the columns "Feature", "type" (binary, real or const), "p_value" (the significance of this feature as a p-value, lower means more significant) "rejected" (if the Benjamini Hochberg procedure rejected this feature) :rtype: pandas.DataFrame """ if settings is None: settings = FeatureSignificanceTestsSettings() target_is_binary = len(set(y)) == 2 # todo: solve the multiclassification case. for a multi classification the algorithm considers the target to be # regression. Instead one could perform a binary one versus all classification. # Only allow entries for which the target is known! y = y.astype(np.float) X = X.copy().loc[~(y == np.NaN), :] # Create the DataFrame df_features containing the information about the different hypotheses # Every row contains information over one feature column from X df_features = pd.DataFrame() # Don't process features from the ignore-list df_features['Feature'] = list(set(X.columns)) df_features = df_features.set_index('Feature', drop=False) # Don't process constant features for feature in df_features['Feature']: if len(pd.unique(X[feature])) == 1: df_features = df_features.drop(feature) _logger.warning( "[test_feature_significance] Feature {} is constant".format( feature)) # Add relevant columns to df_features df_features["type"] = np.nan df_features["p_value"] = np.nan df_features["rejected"] = np.nan # Process the features for feature in df_features['Feature']: if target_is_binary: # Decide if the current feature is binary or not if len(set(X[feature].values)) == 2: df_features.loc[df_features.Feature == feature, "type"] = "binary" p_value = target_binary_feature_binary_test( X[feature], y, settings) else: df_features.loc[df_features.Feature == feature, "type"] = "real" p_value = target_binary_feature_real_test( X[feature], y, settings) else: # Decide if the current feature is binary or not if len(set(X[feature].values)) == 2: df_features.loc[df_features.Feature == feature, "type"] = "binary" p_value = target_real_feature_binary_test( X[feature], y, settings) else: df_features.loc[df_features.Feature == feature, "type"] = "real" p_value = target_real_feature_real_test( X[feature], y, settings) # Add p_values to df_features df_features.loc[df_features['Feature'] == feature, "p_value"] = p_value # Check for constant features for feature in list(set(X.columns)): if len(pd.unique(X[feature])) == 1: df_features.loc[feature, "type"] = "const" df_features.loc[feature, "rejected"] = True # Perform the real feature rejection df_features = benjamini_hochberg_test(df_features, settings) if settings.write_selection_report: # Write results of BH - Test to file if not os.path.exists(settings.result_dir): os.mkdir(settings.result_dir) with open(os.path.join(settings.result_dir, "fs_bh_results.txt"), 'w') as file_out: file_out.write(( "Performed BH Test to control the false discovery rate(FDR); \n" "FDR-Level={0};Hypothesis independent={1}\n").format( settings.fdr_level, settings.hypotheses_independent)) df_features.to_csv(index=False, path_or_buf=file_out, sep=';', float_format='%.4f') return df_features
def test_checks_feature_is_series(self, real_series): with pytest.raises(TypeError): target_real_feature_real_test(x=real_series.values, y=real_series)