[FIX] Fixes 948 multivariate predict/val issues when n_series > 1024

Nixtla / neuralforecast

Scalable and user friendly neural :brain: forecasting algorithms.

https://nixtlaverse.nixtla.io/neuralforecast

Apache License 2.0

2.7k stars 312 forks source link

[FIX] Fixes 948 multivariate predict/val issues when n_series > 1024 #962

Closed elephaint closed 2 months ago

elephaint commented 2 months ago

By default, valid_batch_size = 1024 in BaseModel which gives an issue with BaseMultivariate if n_series > 1024. This fix simply sets the valid_batch_size =n_series, similar to how batch_size is implemented in BaseMultivariate.

Summary:

Fixes #948
Added unit test to TSMixerx model that fails without the fix (unit test should also work for other multivariate models, I only included it for TSMixerx though, since it's more related to BaseMultivariate than a specific model)

review-notebook-app[bot] commented 2 months ago

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.