qolmat.benchmark.metrics.frechet_distance

qolmat.benchmark.metrics.frechet_distance(df1: DataFrame, df2: DataFrame, df_mask: DataFrame, method: str = 'single', min_n_rows: int = 10) Series[source]

Compute Frechet distance computed using a pattern decomposition.

Several variant are implemented: i) the single method relies on a single estimation of the means and covariance matrix. It is relevent for MCAR data. ii) the pattern method relies on the aggregation of the estimated distance between each pattern. It is relevent for MAR data.

Parameters
df1pd.DataFrame

First empirical distribution

df2pd.DataFrame

Second empirical distribution

df_maskpd.DataFrame

Mask indicating on which values the distance has to computed on

method: str

Method used to compute the distance on multivariate datasets with missing values. Possible values are robust and pattern.

min_n_rows: int

Minimum number of rows for a KL estimation

Returns
pd.Series

Series of computed metrics