Hi everyone, I have another question because now I am trying to generate prediction intervals with TimeGPT. When I use the following line of code: timegpt_fcst_df = nixtla_client.forecast(df=df, h=96, freq='15min', id_col='unique_id', level=[10, 20, 30, 40, 50, 60, 70, 80, 90], time_col='ds', target_col='y') I get the following error message:

ValueError: Some series are too short. Please make sure that each serie contains at least 144 observations.

However, when I use the same code without the level-parameter, it works. Is there a length constraint when using the level parameter? Many thanks in advance!

Yes, because at the moment we use conformal prediction to generate levels, which requires a (rather large) minimum number of inputs. We are trying to reduce that number, and have already reduced the minimum size in our latest beta versions.

Okay thank you! How can I use the beta versions? Or is there no beta version where an input length of 96 rows would be enough for all percentile intervals?