ENSTA-U2IS-AI / torch-uncertainty

Open-source framework for uncertainty and deep learning models in PyTorch :seedling:
https://torch-uncertainty.github.io
Apache License 2.0
223 stars 16 forks source link

:bug: Fix the learning rate scale #60

Open o-laurent opened 8 months ago

o-laurent commented 8 months ago

Thanks again to @hanruisong00 for the remark. We have to discuss the best way to fix this scaling problem. The current solution is unsatisfactory as it scales all ensemble methods; however, BatchEnsembles and Masksembles should not be scaled according to their source code. Is the default automated scaling, and are BatchEnsembles/Masksembles anomalies? No, because MCDropout should not be scaled either. If it's 50/50, we could add a parameter to handle this question correctly.

Will resolve #59.

codecov[bot] commented 8 months ago

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Comparison is base (5e67064) 98.39% compared to head (d6df25c) 98.39%.

Additional details and impacted files ```diff @@ Coverage Diff @@ ## dev #60 +/- ## ======================================= Coverage 98.39% 98.39% ======================================= Files 71 71 Lines 3607 3608 +1 Branches 426 426 ======================================= + Hits 3549 3550 +1 Misses 22 22 Partials 36 36 ``` | [Flag](https://app.codecov.io/gh/ENSTA-U2IS/torch-uncertainty/pull/60/flags?src=pr&el=flags&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=ENSTA-U2IS) | Coverage Δ | | |---|---|---| | [cpu](https://app.codecov.io/gh/ENSTA-U2IS/torch-uncertainty/pull/60/flags?src=pr&el=flag&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=ENSTA-U2IS) | `98.39% <100.00%> (+<0.01%)` | :arrow_up: | | [pytest](https://app.codecov.io/gh/ENSTA-U2IS/torch-uncertainty/pull/60/flags?src=pr&el=flag&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=ENSTA-U2IS) | `98.39% <100.00%> (+<0.01%)` | :arrow_up: | Flags with carried forward coverage won't be shown. [Click here](https://docs.codecov.io/docs/carryforward-flags?utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=ENSTA-U2IS#carryforward-flags-in-the-pull-request-comment) to find out more. | [Files](https://app.codecov.io/gh/ENSTA-U2IS/torch-uncertainty/pull/60?src=pr&el=tree&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=ENSTA-U2IS) | Coverage Δ | | |---|---|---| | [torch\_uncertainty/routines/classification.py](https://app.codecov.io/gh/ENSTA-U2IS/torch-uncertainty/pull/60?src=pr&el=tree&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=ENSTA-U2IS#diff-dG9yY2hfdW5jZXJ0YWludHkvcm91dGluZXMvY2xhc3NpZmljYXRpb24ucHk=) | `98.29% <100.00%> (+<0.01%)` | :arrow_up: |

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.