Fix test_pytorch_lightning.py

[Alnusjaponica]

Motivation

Resolve #3418 and #4116

Description of the changes

Refactor deprecated features

• trainer.training_type_plugin is deleted since v1.8 (PR#11239). The attribute training_type_plugin is just renamed to strategy, so refactored as suggested.

• An optional argument of trainer, accelerator stopped to accept ddp_cpu. Instead, we can pass ddp to strategy and cpu to accelerator. Also, num_processes will be removed, so we give the number of processes to devices instead.

• AcceleratorConnector.distributed_backend is deleted, but now they have AcceleratorConnector.is_distributed instead, so refactored as suggested.

• callback.on_init_start() is deleted since v1.8 (Issue#10894, PR#10940, PR#14867).
  Although they don't provide exactly equivalent alternative, it would be possible to move this confirmation to somewhere else. Although it seems strategy.setup_environment is the right place to implement this check, implementing as a method of Strategy will affect user's code. Therefore it will be reasonable to implement as callback.setup or callback.on_fit_start().

[grburgess]

Out of curiosity, is all working for PL 1.8 for this branch w.r.t. the pruner?

[codecov-commenter]

Codecov Report

Merging #4305 (7e88045) into master (7bc4bb1) will decrease coverage by 0.04%.
The diff coverage is 0.00%.

@@            Coverage Diff             @@
##           master    #4305      +/-   ##
==========================================
- Coverage   90.42%   90.38%   -0.04%     
==========================================
  Files         172      172              
  Lines       13660    13668       +8     
==========================================
+ Hits        12352    12354       +2     
- Misses       1308     1314       +6     

Impacted Files	Coverage Δ
optuna/integration/pytorch_lightning.py	0.00% <0.00%> (ø)
optuna/trial/_trial.py	95.26% <0.00%> (-1.19%)	⬇️
optuna/study/study.py	94.96% <0.00%> (+0.77%)	⬆️
optuna/integration/botorch.py	97.55% <0.00%> (+0.81%)	⬆️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[Alnusjaponica]

Thanks for the question. It is what we aim to do in this PR. @grburgess

[grburgess]

Thanks! I checked out this branch to try (which I should have done in the first place) and indeed the issue seems fixed

[Alnusjaponica]

Thank you for trying the patch. Please note that the problem of distributed processing still remains for now.

[Alnusjaponica]

Problem

When you use DDP and optuna.TrialPruned() is raised from the child process, Pytorch Lightning tries to resynchronize to fix the "error" and finally deal it as a DeadlockDetectedException, which terminates the whole process. For more details, see reconciliate_processes. To fix this problem, we need to change environment variable and private variable.

Alternative

Stop supporting DDP in order to fix daily test integration (#4116) and to support PyTorch Lightning after v1.6.

[nzw0301]

When I tried to fix this issue before, I also thought that calling TrialPruned in the main process manually like catboost callback is another approach,
https://github.com/optuna/optuna-examples/blob/main/catboost/catboost_pruning.py#L57.

[Alnusjaponica]

@nzw0301
Thank you for your advice. I'll see if the suggested approach works for this case.
If it is possible, I am going to apply the change to support DDP after #4322 .

[github-actions]

This pull request has not seen any recent activity.

[github-actions]

This pull request was closed automatically because it had not seen any recent activity. If you want to discuss it, you can reopen it freely.