sktime/sktime

[ENH] Allow `ForecastingGridSearchCV` to return results with multiple score metrics

Aberta

#6.012 aberto em 26 de fev. de 2024

 (57 comentários) (0 reação) (1 responsável)Python (1.192 forks)batch import
enhancementfeature requestgood first issuemodule:forecasting

Métricas do repositório

Stars
 (7.162 estrelas)
Métricas de merge de PR
 (Mesclagem média 26d 10h) (86 fundiu PRs em 30d)

Description

Is your feature request related to a problem? Please describe. I would like to have the same dataframe that currently ForecastingGridSearchCV returns as the attribute named cv_results_ but on top of this i would like ForecastingGridSearchCV to evaluate the test set with multiple metrics as different metrics provide different information and currently ForecastingGridSearchCV just returns the metric used to rank the hyperparamters.

Describe the solution you'd like Just return one column per metric in the dataframe. A similar approach is followed by PyCaret, e.g. https://github.com/pycaret/pycaret/blob/master/tutorials/time_series/forecasting/univariate_without_exogeneous_part2.ipynb

Look at the 10th output of the jupyter notebook.

Screenshot 2024-02-26 at 13 56 21

You should be able to pass a list of metrics and then also provide the information of which of the metrics are you going to use to rank the forecaster hyperparams.

Guia do colaborador