Survival pickle - Githubissues

stanfordmlgroup / ngboost

Natural Gradient Boosting for Probabilistic Prediction

Apache License 2.0

1.63k stars 214 forks source link

Survival pickle #246

Closed MikeOMa closed 3 years ago

MikeOMa commented 3 years ago

Made NGBSurvival pickleable. This will not work if the base Dist is not pickleable itself. I assume this would be quite rare though?
Added pickle tests from the test-pickle branch. I hope it gives credit to Ryan as it is all their code! I think I used checkout to copy it directly from the branch.
Testing was taking quite a while so I turned down the number of iterations in the MVN test and the pickle tests
Fixed a small typo in the getstate comments.

Should close https://github.com/stanfordmlgroup/ngboost/issues/238 https://github.com/stanfordmlgroup/ngboost/issues/228

Can change any of it/try to improve it if anyone has any ideas!

MikeOMa commented 3 years ago

Any ideas about why make lint on my local pc gives me different results to github actions? I'll sort the failed checks out tomorrow

MikeOMa commented 3 years ago

Realised putting the class factory SurvivalDistnClass in api is probably a bad idea as people may attempt to use it. Should I move it to distns? Also there was a bug in test pickling introduced by me, fixing it in the next commit

alejandroschuler commented 3 years ago

Realised putting the class factory SurvivalDistnClass in api is probably a bad idea as people may attempt to use it. Should I move it to distns? Also there was a bug in test pickling introduced by me, fixing it in the next commit

You can put it in distns/utils.py or something like that

ryan-wolbeck commented 3 years ago

Any ideas about why make lint on my local pc gives me different results to github actions? I'll sort the failed checks out tomorrow

Are you using make lint? See below https://github.com/stanfordmlgroup/ngboost/pull/204

MikeOMa commented 3 years ago

Any ideas about why make lint on my local pc gives me different results to github actions? I'll sort the failed checks out tomorrow

Are you using make lint? See below

204

Edit: It is 100% a local problem to me. I ran the above on a virtual machine and it had the same as git. Edit2: Deleted ~/.pylintrc and ~/.pylint.d/ and it works as expected now Sorry about the confusion

MikeOMa commented 3 years ago

I may have caused some problems, I fixed a typo in the NGBSurvival class which was also fixed by https://github.com/stanfordmlgroup/ngboost/pull/245

I am pretty the fixes are exactly the same and mine are a subset of #245's fixes.

If this causes any conflicts/issues I can try to fix them but I'm not sure what the best way is!

ryan-wolbeck commented 3 years ago

I may have caused some problems, I fixed a typo in the NGBSurvival class which was also fixed by #245

I am pretty the fixes are exactly the same and mine are a subset of #245's fixes.

If this causes any conflicts/issues I can try to fix them but I'm not sure what the best way is!

I'm not seeing any conflicts so I think we should be good

ryan-wolbeck commented 3 years ago

Any ideas about why make lint on my local pc gives me different results to github actions? I'll sort the failed checks out tomorrow

Are you using make lint? See below

204

Edit: It is 100% a local problem to me. I ran the above on a virtual machine and it had the same as git. Edit2: Deleted ~/.pylintrc and ~/.pylint.d/ and it works as expected now Sorry about the confusion

Awesome! I'd recommend running everything in a virtualenv

ryan-wolbeck commented 3 years ago

@MikeOMa thanks for doing this! I'm bumping the version then will merge after the tests pass