AutoSeasonalityAndDifferences crashes the kernel under certain scenarios

[elephaint]

What happened + What you expected to happen

The following code causes a kernel crash on Ubuntu (WSL) and Windows 11, both Python 3.11, on a fresh uv environment.

The kernel doesn't crash when reducing max_diffs inside AutoSeasonalityAndDifferences to 4. I'm guessing an illegal memory issue (ptr to illegal/unexisting memory address).

Probably certain combinations of parameters shouldn't be allowed, i.e. many differences with seasonality of 12 isn't possible if the shortest series is of length 15 (as in below example).

Versions / Dependencies

• Python 3.11
• Ubuntu 22.04 WSL
• Windows 11 Pro

Reproduction script

import lightgbm as lgb
from xgboost import XGBRegressor
from mlforecast import MLForecast
from mlforecast.lag_transforms import ExpandingMean, RollingMean,ExponentiallyWeightedMean
from mlforecast.target_transforms import AutoSeasonalityAndDifferences, LocalStandardScaler
from utilsforecast.data import generate_series

series_count=100
params={
  "model_frequency":"MS",
  "model_horizon":12,
  "season":12,
  "eval_interval":3
}

training_data = generate_series(series_count, min_length=15, max_length=120, freq=params['model_frequency'], with_trend=True, static_as_categorical=False)

ml_models_list = [
lgb.LGBMRegressor(random_state=0, verbosity=-1),
XGBRegressor( n_estimators=500, learning_rate=0.05)
]
date_features = ['month']

ml = MLForecast(
    models=ml_models_list,
    freq=params['model_frequency'],
    lags=[1, 3],
    lag_transforms={
    1: [ExpandingMean()],
    12: [RollingMean(window_size=params['model_horizon'])],
    12: [ExponentiallyWeightedMean(alpha=0.4)],
    },
    date_features= date_features,
    target_transforms=[ LocalStandardScaler (),
              AutoSeasonalityAndDifferences(max_diffs=12,
                            max_season_length=params['season'],
                            n_seasons=3)
              ]
    )
ml.fit(training_data)

Issue Severity

None