Problem with statsforecast mstl_decomposition

[Jonathan-87]

What happened + What you expected to happen

When using mstl_decomposition if I'm passing a multiple frequencies like 7 and 365 for daily data then I'm getting an error saying "KeyError: "['seasonal'] not in index"" because when applying multiple frequencies the name of the resulting variables will be seasonal7 and seasonal365

Versions / Dependencies

statsforecast 1.7.5

Reproduction script

from statsforecast import StatsForecast
from statsforecast.feature_engineering import mstl_decomposition
from statsforecast.models import MSTL, AutoARIMA

df = pd.read_csv('https://datasets-nixtla.s3.amazonaws.com/air-passengers.csv', parse_dates=['ds'])

#Here i used 12 and 24 for season length just for demonstration
model = MSTL(season_length=[12, 24], trend_forecaster=AutoARIMA())
df_feat, df_fut = mstl_decomposition(df, model=model, freq="M", h=horizon)

Issue Severity

None

[jmoralez]

Hey @Jonathan-87, thanks for raising this. The current implementation leverages a function that adds up the seasonal components into one, so the easiest fix would be to always add up the seasonal components, however I realize that it may be better to have them as individual features. What would be the expected output for you? Generating a trend + one feature for each seasonal component or just trend + seasonal (where the seasonal is the sum of all components)?

[Jonathan-87]

Hi @jmoralez I think it's better to have the different seasonnalities as individual features.