[Model] Sample weights do not work in mlf.cross_validation with refit=False

[slamao]

Sample weights do not work in mlf.cross_validation with refit=False

Hi,

I found out that mlforecast's cross validation method does not work in one specific case: when using sample weights and when refit=False. I think that this is a correct setting and that the cross validation should work in this case as well (please correct me if I am wrong).

There is a reproducible example below:

from mlforecast.utils import generate_daily_series
from mlforecast import MLForecast
from sklearn.linear_model import ElasticNet

series = generate_daily_series(2)
series['weight'] = np.random.default_rng(seed=0).random(series.shape[0])
series['exog'] = np.random.default_rng(seed=0).random(series.shape[0])

fcst = MLForecast(
    models={
        'lr': ElasticNet(alpha=0.1)
    },
    freq='D',
    lags=[1],
    date_features=['dayofweek'],
)

# cross validation throws error
fcst.cross_validation(series, weight_col='weight', static_features=[], n_windows=2, h=7, refit=False)

Similar error is also thrown when not using any exogenous variables.

Versions / Dependencies

python: 3.12
mlforecast: 1.0.1
pandas: 2.2.3
scikit-learn: 1.6.1

Reproduction script

from mlforecast.utils import generate_daily_series
from mlforecast import MLForecast
from sklearn.linear_model import ElasticNet

series = generate_daily_series(2)
series['weight'] = np.random.default_rng(seed=0).random(series.shape[0])
series['exog'] = np.random.default_rng(seed=0).random(series.shape[0])

fcst = MLForecast(
    models={
        'lr': ElasticNet(alpha=0.1)
    },
    freq='D',
    lags=[1],
    date_features=['dayofweek'],
)

# cross validation throws error
fcst.cross_validation(series, weight_col='weight', static_features=[], n_windows=2, h=7, refit=False)

Issue Severity

Medium: It is a significant difficulty but I can work around it.