Window size of composed pipeline can be too large
The snippet
auto_model = Fedot(problem=problem, timeout=0.2, task_params=TsForecastingParams(forecast_length=14), composer_params={'metric': 'rmse'}, ) pipeline = auto_model.fit(features=features, target=target)
max_lag = \ round(max([np.ceil(n.custom_params['window_size']) for n in pipeline.nodes if str(n.operation) == 'lagged']))
can produce lags with higher length then entire dataset length. It make it impossible to estimate the required length of pre-history.