orbit/template/dlt.py [819:909]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - seasonal_component = seasonal_component[:, start:] # sum components pred_array = trend_component + seasonal_component + regression pred_array = pred_array.numpy() trend_component = trend_component.numpy() seasonal_component = seasonal_component.numpy() regression = regression.numpy() out = { PredictionKeys.PREDICTION.value: pred_array, PredictionKeys.TREND.value: trend_component, PredictionKeys.SEASONALITY.value: seasonal_component, PredictionKeys.REGRESSION.value: regression, } return out def get_regression_coefs( self, training_meta, point_method, point_posteriors, posterior_samples, lower=0.05, upper=0.95, ): """Return DataFrame regression coefficients. If point_method is None when fitting, return the median of coefficients. Parameters ----------- lower : float between (0, 1). default to be 0.05 lower bound for the CI upper : float between (0, 1). default to be 0.95. upper bound for the CI Returns ------- pandas data frame holding the regression coefficients """ # init dataframe coef_df = pd.DataFrame() # end if no regressors if self.num_of_regressors == 0: return coef_df _point_method = point_method if point_method is None: _point_method = PredictMethod.MEDIAN.value coef = point_posteriors.get(_point_method).get( RegressionSamplingParameters.REGRESSION_COEFFICIENTS.value ) # get column names pr_cols = self.positive_regressor_col nr_cols = self.negative_regressor_col rr_cols = self.regular_regressor_col # note ordering here is not the same as `self.regressor_cols` because regressors here are grouped by signs regressor_cols = pr_cols + nr_cols + rr_cols # same note regressor_signs = ( ["Positive"] * self.num_of_positive_regressors + ["Negative"] * self.num_of_negative_regressors + ["Regular"] * self.num_of_regular_regressors ) coef_df[COEFFICIENT_DF_COLS.REGRESSOR] = regressor_cols coef_df[COEFFICIENT_DF_COLS.REGRESSOR_SIGN] = regressor_signs coef_df[COEFFICIENT_DF_COLS.COEFFICIENT] = coef.flatten() # if we have posteriors distribution and also include ci if point_method in [None, PredictMethod.MEAN.value, PredictMethod.MEDIAN.value]: coef_samples = posterior_samples.get( RegressionSamplingParameters.REGRESSION_COEFFICIENTS.value ) coef_lower = np.quantile(coef_samples, lower, axis=0) coef_upper = np.quantile(coef_samples, upper, axis=0) coef_df[COEFFICIENT_DF_COLS.COEFFICIENT + "_lower"] = coef_lower coef_df[COEFFICIENT_DF_COLS.COEFFICIENT + "_upper"] = coef_upper n_pos = np.apply_along_axis(lambda x: np.sum(x >= 0), 0, coef_samples) n_total = coef_samples.shape[0] coef_df[COEFFICIENT_DF_COLS.PROB_COEF_POS] = n_pos / n_total coef_df[COEFFICIENT_DF_COLS.PROB_COEF_NEG] = 1 - n_pos / n_total return coef_df - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - orbit/template/lgt.py [686:776]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - seasonal_component = seasonal_component[:, start:] # sum components pred_array = trend_component + seasonal_component + regression pred_array = pred_array.numpy() trend_component = trend_component.numpy() seasonal_component = seasonal_component.numpy() regression = regression.numpy() out = { PredictionKeys.PREDICTION.value: pred_array, PredictionKeys.TREND.value: trend_component, PredictionKeys.SEASONALITY.value: seasonal_component, PredictionKeys.REGRESSION.value: regression, } return out def get_regression_coefs( self, training_meta, point_method, point_posteriors, posterior_samples, lower=0.05, upper=0.95, ): """Return DataFrame regression coefficients. If point_method is None when fitting, return the median of coefficients. Parameters ----------- lower : float between (0, 1). default to be 0.05 lower bound for the CI upper : float between (0, 1). default to be 0.95. upper bound for the CI Returns ------- pandas data frame holding the regression coefficients """ # init dataframe coef_df = pd.DataFrame() # end if no regressors if self.num_of_regressors == 0: return coef_df _point_method = point_method if point_method is None: _point_method = PredictMethod.MEDIAN.value coef = point_posteriors.get(_point_method).get( RegressionSamplingParameters.REGRESSION_COEFFICIENTS.value ) # get column names pr_cols = self.positive_regressor_col nr_cols = self.negative_regressor_col rr_cols = self.regular_regressor_col # note ordering here is not the same as `self.regressor_cols` because regressors here are grouped by signs regressor_cols = pr_cols + nr_cols + rr_cols # same note regressor_signs = ( ["Positive"] * self.num_of_positive_regressors + ["Negative"] * self.num_of_negative_regressors + ["Regular"] * self.num_of_regular_regressors ) coef_df[COEFFICIENT_DF_COLS.REGRESSOR] = regressor_cols coef_df[COEFFICIENT_DF_COLS.REGRESSOR_SIGN] = regressor_signs coef_df[COEFFICIENT_DF_COLS.COEFFICIENT] = coef.flatten() # if we have posteriors distribution and also include ci if point_method in [None, PredictMethod.MEAN.value, PredictMethod.MEDIAN.value]: coef_samples = posterior_samples.get( RegressionSamplingParameters.REGRESSION_COEFFICIENTS.value ) coef_lower = np.quantile(coef_samples, lower, axis=0) coef_upper = np.quantile(coef_samples, upper, axis=0) coef_df[COEFFICIENT_DF_COLS.COEFFICIENT + "_lower"] = coef_lower coef_df[COEFFICIENT_DF_COLS.COEFFICIENT + "_upper"] = coef_upper n_pos = np.apply_along_axis(lambda x: np.sum(x >= 0), 0, coef_samples) n_total = coef_samples.shape[0] coef_df[COEFFICIENT_DF_COLS.PROB_COEF_POS] = n_pos / n_total coef_df[COEFFICIENT_DF_COLS.PROB_COEF_NEG] = 1 - n_pos / n_total return coef_df - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -