Hi, I've been encountering this error. Some of my series have less than 27 samples but most of them have more. I've been running the same model for different levels of hierarchy ie store, subcategory, product and weirdly this only happens at the subcategory level even though there are far more at a product level that have less than 27 sample than at a subcategory level. Subcategory level has less samples overall though so could it be looking at the it as a percentage of samples less than 27? @José Morales any idea what's going on?

If some have less than 27 samples, the error indicates that that is not allowed. Did you try reducing n_windows?

Hi @Olivier yeah but this doesn't appear to be consistent with my dataset. For example when I run a dataset that has far more series with less than 27 samples it runs fine. Is there anyway for series with less to just be ignored rather than it causing a failure?

You can delete those series beforehand from your data? Then they will be ignored.... This might be a genuine bug and should be simple to reproduce; if you generate synthetic series with samples ranging between e.g. 10 to 50 samples and then use your training script the error should pop up if it's a genuine bug. You can use utilsforecast to generate synthetic data to construct a reproducible example. Post it here and I can have a look to further investigate.