Loss functions module #640
Merged
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
Bumps [sqlparse](https://github.com/andialbrecht/sqlparse) from 0.4.4 to 0.5.0. - [Changelog](https://github.com/andialbrecht/sqlparse/blob/master/CHANGELOG) - [Commits](andialbrecht/sqlparse@0.4.4...0.5.0) --- updated-dependencies: - dependency-name: sqlparse dependency-type: indirect ... Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Bumps [tqdm](https://github.com/tqdm/tqdm) from 4.66.1 to 4.66.3. - [Release notes](https://github.com/tqdm/tqdm/releases) - [Commits](tqdm/tqdm@v4.66.1...v4.66.3) --- updated-dependencies: - dependency-name: tqdm dependency-type: indirect ... Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Bumps [werkzeug](https://github.com/pallets/werkzeug) from 3.0.1 to 3.0.3. - [Release notes](https://github.com/pallets/werkzeug/releases) - [Changelog](https://github.com/pallets/werkzeug/blob/main/CHANGES.rst) - [Commits](pallets/werkzeug@3.0.1...3.0.3) --- updated-dependencies: - dependency-name: werkzeug dependency-type: indirect ... Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Bumps [jinja2](https://github.com/pallets/jinja) from 3.1.3 to 3.1.4. - [Release notes](https://github.com/pallets/jinja/releases) - [Changelog](https://github.com/pallets/jinja/blob/main/CHANGES.rst) - [Commits](pallets/jinja@3.1.3...3.1.4) --- updated-dependencies: - dependency-name: jinja2 dependency-type: indirect ... Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Bumps [mlflow](https://github.com/mlflow/mlflow) from 2.10.1 to 2.12.1. - [Release notes](https://github.com/mlflow/mlflow/releases) - [Changelog](https://github.com/mlflow/mlflow/blob/master/CHANGELOG.md) - [Commits](mlflow/mlflow@v2.10.1...v2.12.1) --- updated-dependencies: - dependency-name: mlflow dependency-type: direct:development ... Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Bumps [gunicorn](https://github.com/benoitc/gunicorn) from 21.2.0 to 22.0.0. - [Release notes](https://github.com/benoitc/gunicorn/releases) - [Commits](benoitc/gunicorn@21.2.0...22.0.0) --- updated-dependencies: - dependency-name: gunicorn dependency-type: indirect ... Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
updated-dependencies: - dependency-name: requests dependency-type: indirect ... Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
* try a simple workflow first * try running on new ubuntu VM * fixes * bump poetry version to 1.8.3 * try removing caching.. * add workflow for testing tsv tools
* try skipping test_tsvtools when PR is in draft mode * trigger CI * add a cpu tag to avoid running cpu tests on gpu machines * run also on refactoring branch
* add test workflow on GPU for train * fix conda path * fix conflicting workdir * only run on non-draft PRs * run also on refactoring branch
* add workflow for testing interpretation task * add workflow for testing random search task * add workflow for testing resume task * add workflow for testing transfer learning task * trigger CI * trigger CI
* add cleaning step to test_tsvtools pipeline * add test_generate pipeline * add test_predict pipeline * add test_prepare_data pipeline * add test_quality_checks pipeline * add refactoring target branch, cpu tag, and draft PR filter * trigger CI
…nt in README (aramis-lab#600) * update python version used for creating conda env in README * investigate * fix
* add no-gpu and adapt-base-dir flag
* Update quality_check.py
* add FileNotFound error in tree
|
Update: I added an option "DefaultFromLibrary" in the config class to use by default the default values from PyTorch's loss functions, instead of manually copy/paste these defaults in our config class. |
clinicadl/losses/factory.py
Outdated
| return loss, config_dict | ||
|
|
||
|
|
||
| # TODO : what about them? |
There was a problem hiding this comment.
Are you talking about the other losses?
I've just added more losses from PyTorch in my last commit.
| if "weight" in config_dict and config_dict["weight"] is not None: | ||
| config_dict_["weight"] = torch.Tensor(config_dict_["weight"]) | ||
| loss = loss_class(**config_dict_) | ||
|
|
There was a problem hiding this comment.
As the loss is already initialized here, do we need to return a dictionary? Because I thought we wanted to get rid of all the dictionaries.
There was a problem hiding this comment.
Hi @camillebrianceau, thanks for the review. You're right, I changed that to put the arguments of the loss function in a config class, rather than a dict.
Even if the loss is initialized, we need the config class to store the config in the MAPS.
The difference between the input and the output config class is that the values set to "DefaultFromLibrary" have been changed with their effective values. It is not essential but better for reproducibility I think (for example in the - unlikely - event that PyTorch changes some default values).
Tell me what do you think!
Merged
Merged
Merged
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.
In this PR, I suggest a new module for loss functions. The module is very simple and is actually only an intermediary between
ClinicaDLand PyTorch'snnmodule. More precisely, there is:LossConfigclass, with all the parameters useful for PyTorch's loss functions,get_loss_functionthat will get a PyTorch loss function and parametrize it thanks to aLossConfigobject. This function also returns a config dict with only the parameters relevant to the wanted loss function (whereas aLossFunctionobject contains the parameters for all loss functions available), because one only wants to save the information related to the wanted loss function in the MAPS.