evidentlyai / evidently

Evidently is ​​an open-source ML and LLM observability framework. Evaluate, test, and monitor any AI-powered system or data pipeline. From tabular data to Gen AI. 100+ metrics.
https://www.evidentlyai.com/evidently-oss
Apache License 2.0
5.2k stars 586 forks source link

Missing "Per feature"-metrics in Regression #1075

Closed DonnyWhoLovedBowling closed 5 months ago

DonnyWhoLovedBowling commented 5 months ago

Somehow I can't replicatete full regression performance report as it is shown on: https://docs.evidentlyai.com/presets/reg-performance

All the "per-feature" analyses plots seem to be missing (components 10-12 in the above link)

I ran this notebook (as a norml python-script): https://github.com/evidentlyai/community-examples/blob/main/tutorials/bicycle_demand_monitoring_setup.ipynb

And still all the "per-feature" components seem to be missing. I don't see any warnings or errors occurring and I updated evidently to several versions, including the latest (0.4.19) , so currently I am a bit in the dark on what's going on.

Thanks in advance!

Edit: by debugging/digging in the evidently code I found that the problem was that the option "render.raw_data" needed to be set to True. I've got the plots I want, but it might be a good idea to document this somewhere or explain in the tutorials section why this is necessary.

elenasamuylova commented 5 months ago

Hi @DonnyWhoLovedBowling,

You found the correct solution, this visualization requires passing the raw data option.

report = Report(
    metrics=[
      ClassificationQualityByFeatureTable(),
    ],
    options={"render": {"raw_data": True}}
  )
report.run(reference_data=ref, current_data=cur)
report

It is in the backlog to re-implement this metric to support the non-aggregated visualization. (Either this or the documentation update is coming soon :) You are also welcome to submit a PR to the docs to add a note about this!

DonnyWhoLovedBowling commented 5 months ago

Hi @elenasamuylova,

thanks for your quick reply! I just noticed that if you search your website on "raw data", you actually find the right documentation (https://docs.evidentlyai.com/user-guide/customization/report-data-aggregation#non-aggregated-plots-for-reports)

Perhaps it might be an easy solution just to place a link to this page at places where it is relevant. For example: https://docs.evidentlyai.com/presets/reg-performance Or it might be nice for completeness/overview to add a mention in this table for each metric where it is relevant? https://docs.evidentlyai.com/reference/all-metrics

For now I'm closing this issue, thanks again.