ARIMARegressor should fill boolean nans with the mode then convert to double to fully support boolean values
- Lingua principale
- Python
- Stelle
- 850
- Fork
- 96
- Metriche di merge delle PR
- Nessuna PR unita negli ultimi 30g
Descrizione
- As a user, I wish I could pass any boolean column into the ARIMARegressor component. Currently, if you pass in a column with the `Boolean` logical type, we convert it to `Double` prior to passing into the estimator. But we do nothing with `BooleanNullable` columns, which will cause them to raise an error that AutoARIMA can't handle boolean values.
We should add support for all boolean values. Note - this is NOT a nullable type incompatibility. Sktime's AutoARIMA regressor is not meant to take in any boolean value -- nullable type or not--and we do not expect them to add support for that. It is our choice in evalml to convert booleans to doubles to pass into ARIMA, so we should support all boolean types.
We'll need to handle potential null values in the same way we handle null values in numeric columns but use the mode instead of the mean. This means we should `fillna` with the mode of the boolean nullable columns, then convert to Double.
Guida per i contributori
Apri la guida per i contributori
Valutazione
Questa issue non è ancora stata valutata.