This page lists known issues with Prevision.io, along with ways you can avoid or recover from these issues. Those issues are given a high priority to the Prevision.io development team and will be solved in an upcoming release.
- There can be small differences between the unit and batch predict for the same data and model.
- The decision tree simple model will not accurately display the modalities for categorical features, instead only showing the encoded modalities. (red, green, blue will show up as 1, 2, 3)
- Downloading large files (> 1GB) from the notebooks can sometimes fail silently (the downloaded file might be incomplete).
- For multi-classification usecases, the bivariate analysis shows the top 3 classes per bin instead of the top 3 classes overall.
- The .svg format plot download will sometimes fail.
- Dataset upload will fail with empty datasets (only header). In that case the « Parsed » status will never complete or fail.
- Some advanced metrics (F2, F3, F4 Score, Lift, …) don’t have a star system related to them
- ALN timeseries usecases might crash when given an ID column