AttributeError: 'Index' object has no attribute 'dt' when passing date features into AutoMLForecast.

[micheldearaujo]

Hey all. I having been trying with Nixtla's libs for a few days and I enjoying it a lot!

I am trying to perform a hyperparameter tuning with some Boosting models with the final objective of doing hierarquical focasting. I am facing an issue when trying to add my features into the AutoMLForecast. Right now I didn't want to perform tuning on the features, just use a small set that I choose, but I was not able to find how to add the features "statically", so I added them into the features tuning like this:

def my_init_config(trial: optuna.Trial):

    lag_transforms = [
        ExponentiallyWeightedMean(alpha=0.3),
        RollingMean(window_size=24 * 7, min_samples=1),
    ]
    lag_to_transform = trial.suggest_categorical('lag_to_transform', [1])

    return {
        'date_features': ['dayofyear'],
        'lags': [1, 7],  # this won't be tuned
        'lag_transforms': {lag_to_transform: lag_transforms}, 
    }

auto_mlf = AutoMLForecast(
    models=[AutoXGBoost()],
    freq=1,
    season_length=24,
    init_config=my_init_config,
).fit(
    train,
    n_windows=2,
    h=horizon,
    num_samples=2,
)

But I am receiving the following error about the date features:

364 feat_vals = getattr(dates, feature)
365 else:
--> 366 feat_vals = getattr(dates.dt, feature)()
367 if isinstance(feat_vals, (pd.Index, pd.Series)):
368 feat_vals = np.asarray(feat_vals)

AttributeError: 'Index' object has no attribute 'dt'.

I tested with multiple date features and the error continues. How can I solve that?

[jmoralez]

Hey. Thanks for using mlforecast. The problem is that your time column has integers, so it doesn't have a dayofyear attribute, that only works if your time column is made of timestamps.

[micheldearaujo]

Oh, I forget that the dataset wasn't using timestamps. Thanks a lot!