Predicting a subset of ids

Compute predictions for only a subset of the training ids

from lightgbm import LGBMRegressor
from fastcore.test import test_fail

from mlforecast import MLForecast
from mlforecast.utils import generate_daily_series

series = generate_daily_series(5)
fcst = MLForecast({'lgb': LGBMRegressor(verbosity=-1)}, freq='D', date_features=['dayofweek'])
fcst.fit(series)
all_preds = fcst.predict(1)
all_preds

	unique_id	ds	lgb
0	id_0	2000-08-10	3.728396
1	id_1	2000-04-07	4.749133
2	id_2	2000-06-16	4.749133
3	id_3	2000-08-30	2.758949
4	id_4	2001-01-08	3.331394

By default all series seen during training will be forecasted with the predict method. If you’re only interested in predicting a couple of them you can use the ids argument.

fcst.predict(1, ids=['id_0', 'id_4'])

	unique_id	ds	lgb
0	id_0	2000-08-10	3.728396
1	id_4	2001-01-08	3.331394

Note that the ids must’ve been seen during training, if you try to predict an id that wasn’t there you’ll get an error.

test_fail(lambda: fcst.predict(1, ids=['fake_id']), contains='fake_id')

Predict callbacks Transfer Learning | MLForecast

⌘I