🔗 Helpful links

• 🧪  See artifacts and rendered test results at hud.pytorch.org/pr/82769
• 📄  Preview Python docs built from this PR
• 📄  Preview C++ docs built from this PR
• ❓Need help or want to give feedback on the CI? Visit our office hours

✅ No Failures (0 Pending)

As of commit 20f6c84 (more details on the Dr. CI page):

@datumbox

First draft. Documentation should be quite ok, but building documentation phase is missing.

Tests can be improved (combined versions are missing, e.g. PolynomialLR + StepLR).

I tried to generate the .pyi for optimizer.py but the result I obtained is quite different from the existing version. This is what I did, using mypy:

stubgen torch/optim/lr_scheduler.py

Generated file:

class PolynomialLR(_LRScheduler):
    total_iters: Incomplete
    min_lrs: Incomplete
    power: Incomplete
    def __init__(self, optimizer, total_iters: int = ..., min_lr: float = ..., power: float = ..., last_epoch: int = ..., verbose: bool = ...) -> None: ...
    def get_lr(self): ...

class LinearLR(_LRScheduler):
    start_factor: Incomplete
    end_factor: Incomplete
    total_iters: Incomplete
    def __init__(self, optimizer, start_factor=..., end_factor: float = ..., total_iters: int = ..., last_epoch: int = ..., verbose: bool = ...) -> None: ...
    def get_lr(self): ...

Existing optimizer.pyi:

class LinearLR(_LRScheduler):
    start_factor: float = ...
    end_factor: float = ...
    total_iters: int = ...
    def __init__(self, optimizer: Optimizer, start_factor: float=..., end_factor: float= ..., total_iters: int= ..., last_epoch: int= ..., verbose: bool = ...) -> None: ...

Note2: I confused torchvision issue with torch one, so the issue number in the branch name is wrong. Should I delete the branch?

I think overall it looks good. I've added only one minor nit comment below.

The question around the generation of optimizer.pyi remains, but we can ask core devs how to handle it. I've restarted the unit-tests so that we can get a signal from the CI. If all tests pass, from my perspective, we should mark the PR as non draft and ask for the review of Alban/Joel.

Some tests are still failing, but it seems that they are not related to my changes... can you check you too, please?

@federicopozzi33 LGTM! The failing tests are unrelated.

I think now we should be good to ask for the feedback of @albanD and @jbschlosser. Please let us know your thoughts and your recommendation for the generation of optimizer.pyi.

The .pyi files are written by hand if they are checked into the git repo. That's why it doesn't match what mypy would generate.

@pytorchbot rebase

@pytorchbot successfully started a rebase job. Check the current status here

Successfully rebased feature/4438-poly-lr-scheduler onto refs/remotes/origin/master, please pull locally before adding more changes (for example, via git checkout feature/4438-poly-lr-scheduler && git pull --rebase)

@pytorchbot merge -g

@pytorchbot successfully started a merge job. Check the current status here

Hey @federicopozzi33.
You've committed this PR, but it does not have both a 'release notes: ...' and 'topics: ...' label. Please add one of each to the PR. The 'release notes: ...' label should represent the part of PyTorch that this PR changes (fx, autograd, distributed, etc) and the 'topics: ...' label should represent the kind of PR it is (not user facing, new feature, bug fix, perf improvement, etc). The list of valid labels can be found here for the 'release notes: ...' and here for the 'topics: ...'.
For changes that are 'topic: not user facing' there is no need for a release notes label.