Report CycleGAN validation metrics correctly to wandb #2131
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
mcgibbon
commented
Jan 13, 2023
.pre-commit-config.yaml
Outdated
| @@ -1,5 +1,10 @@ | |||
| exclude: "external/gcsfs/" | |||
| repos: | |||
| - repo: https://github.com/psf/black | |||
There was a problem hiding this comment.
I'd like black to run before flake8 so that it can auto-fix flake8 issues before flake8 runs.
| generator | ||
| discriminator_optimizer: configuration for the optimizer used to train the | ||
| discriminator | ||
| optimizer: configuration for the optimizer used to train the |
There was a problem hiding this comment.
Merging these was necessary so a wandb sweep can operate on the learning rate, there's no way to pair two hyperparameters for wandb sweeps.
| @@ -316,76 +314,8 @@ def _init_targets(self, shape: Tuple[int, ...]): | |||
| torch.Tensor(shape).fill_(0.0).to(DEVICE), requires_grad=False | |||
| ) | |||
|
|
|||
| def evaluate_on_dataset( | |||
There was a problem hiding this comment.
This function was never actually used (it was called, but only on validation data, and I never provided validation datasets before now).
| @@ -395,6 +325,8 @@ def train_on_batch( | |||
| [sample, time, tile, channel, y, x] | |||
| real_b: a batch of data from domain B, should have shape | |||
| [sample, time, tile, channel, y, x] | |||
| training: if True, the model will be trained, otherwise we will | |||
There was a problem hiding this comment.
This allows getting training metrics on validation data.
| @@ -51,12 +50,11 @@ class CycleGANNetworkConfig: | |||
| cycle_weight: weight of the cycle loss | |||
| generator_weight: weight of the generator's gan loss | |||
| discriminator_weight: weight of the discriminator gan loss | |||
| reload_path: path to a directory containing a saved CycleGAN model to use | |||
There was a problem hiding this comment.
This was just a missing docstring entry.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Currently the CycleGAN training routine does not report the same training losses on validation data. This PR refactors the code to use training losses for validation data, produce one wandb report per epoch, and adds regularization loss as an output metric.
Refactored public API:
Significant internal changes:
Coverage reports (updated automatically):