Open hwoarang09 opened 1 week ago
Hi @hwoarang09,
It seems that lightgbm only allows a maximum value of 255 bins when working with categorical features.
In skforecast >= 0.13.0 we introduced an encoding
argument. Could you try using encoding='ordinal'?
You can see an example here: https://skforecast.org/0.12.1/user_guides/independent-multi-time-series-forecasting#series-encoding-in-multi-series
forecaster = ForecasterAutoregMultiSeries(
regressor = RandomForestRegressor(random_state=123),
lags = 3,
encoding = 'ordinal'
)
i'm studying with you website, https://cienciadedatos.net/documentos/py53-global-forecasting-models.html
there are more than 1000 buildings.
when i resample building with
==============================================================================
end_train = '2016-07-31 23:59:00' end_validation = '2016-09-30 23:59:00' data_train = data.loc[: end_train, :].copy() data_val = data.loc[end_train:end_validation, :].copy() data_test = data.loc[end_validation:, :].copy()
Sample 600 buildings for GPU
==============================================================================
rng = np.random.default_rng(12345) buildings = data['building_id'].unique() buildings_selected = rng.choice( buildings, size = 254, replace = False )
size less than 255 it works.
but if size over 255,
error occur LightGBMError: bin size 257 cannot run on GPU
so i have to resample less than 255...
but i want my model learn all buildings.
Can you help me...?? Thank you!!