Nixtla's Open Source Time Series Ecosystem.

Nixtla Community

hi!

thanks for letting us know about the problem. We are thinking of adding a functionality to the pipeline to consider a “safe model” (e.g. SeasonalNaive) to prevent this kind of errors.

We’ve experienced the `could not fit model`  problem working with `ETS`  and intermittent time series, it this your case? You can find a solution here: <https://github.com/Nixtla/statsforecast/issues/182>

Yeah I kind of assumed that the zeros was the problem and added a small smoothing value - is this the only way to reach this particular error though?

I think even just returning NaNs for series where a model could be fitted would be enough. If I'm using a `Forecast` object with a bunch of models, and one of them fails on one series, it seems a bit annoying to kill the whole pipeline