Skip to content
This repository was archived by the owner on Mar 26, 2025. It is now read-only.

Should the row_filter be applied in the prediction client? #965

Open
flikka opened this issue Mar 16, 2020 · 1 comment
Open

Should the row_filter be applied in the prediction client? #965

flikka opened this issue Mar 16, 2020 · 1 comment

Comments

@flikka
Copy link
Contributor

flikka commented Mar 16, 2020

config = machine.dataset.to_dict()

Previously, the row_filter was not used when doing predictions. After 994a9f8#diff-e7f1b0bbb4d76747e62b1fb5d978a1feL505 I believe it is.

The argument for not using the filter at prediction time is that it is not clear what the row_filter is used for. It is a generic feature to cut data with pandas filters. Doing this during prediction time seems harsh. One could for example intentionally filter out data to build a certain model (low rpm model?), with the precise intention to see how it behaves under other circumstances.

@epa095
Copy link
Contributor

epa095 commented Mar 16, 2020

something should definitely change, just silently dropping the data is clearly bad. And my guess is that it probably screws up some date alignment as well since the server expects data without gaps

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants