# Copyright (c) 2021. Bram Vonk, Enexis import logging import math import numpy as np import pandas as pd [docs]logger = logging.getLogger("SPARK") [docs]def make_quantile_bands(df_base, samples, quantiles=(5, 15, 50, 85, 95)): """ Translate samples to bands with edges defined by the given quantiles and merge them to a base DataFrame. Parameters ---------- df_base : pd.DataFrame base DataFrame with timestamps/length in accordance of the samples shape samples : np.array the posterior predictive samples quantiles: list an iterable with the edges of the bands (0,1), ordered increasingly. Returns ------- pd.DataFrame long format version with the quantile bands of the samples """ # get quantiles q_data = np.quantile(samples, [q / 100 for q in quantiles], axis=0) boundaries = ["upper", "lower"] df_bands = pd.DataFrame() # create bands from two quantile boundaries (median: upper=lower) for ci in range(math.ceil(len(quantiles) / 2)): # name band band_range = f"Q{quantiles[ci]}-Q{quantiles[-ci-1]}".replace("Q50-Q50", "median") df_band = df_base.copy() df_band[boundaries] = q_data[[-ci - 1, ci]].T df_band["band"] = band_range # append to other bands df_bands = pd.concat([df_bands, df_band], axis=0) # in long format df_bands = df_bands.melt( id_vars=df_bands.columns.difference(boundaries), value_vars=boundaries, var_name="boundary", value_name="value", ) return df_bands [docs]def format_model_estimates(df_base, pp, quantiles=(5, 15, 50, 85, 95)): """ Format the samples into quantile bands for every model variable. Parameters ---------- df_base : pd.DataFrame base DataFrame with timestamps/length in accordance of the samples shape pp : dict per model variable (key) the posterior predictive samples quantiles: list an iterable with the edges of the bands (0,1), ordered increasingly. Returns ------- pd.DataFrame long format version with the quantile bands of the samples for every model variable """ logger.info(f"calculating bands for quantiles: {quantiles}") df_vars = pd.DataFrame() for var, samples in pp.items(): # get per model variable the quantile bands logger.info(f"calculating bands for variable: {var}") df_var = make_quantile_bands(df_base, samples=samples, quantiles=(5, 15, 50, 85, 95)).assign(model_var=var) df_vars = pd.concat([df_vars, df_var], axis=0) return df_vars