Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

ENH: Add lightning.Trainer config, fix #691 #687 #742 #745 #752

Merged
merged 47 commits into from
May 5, 2024
Merged

ENH: Add lightning.Trainer config, fix #691 #687 #742 #745 #752

merged 47 commits into from
May 5, 2024

Conversation

NickleDave
Copy link
Collaborator

This adds a trainer table to the train/eval/predict/learncurve tables in the configuration file, with keys accelerator and devices, like so:

[vak.train.trainer]
accelerator = 'gpu'
devices = [0]

Those keys gets passed directly to lightning.Trainer as the corresponding parameters when instantiating. For now we only allow devices to be a list of a single int when accelerator is anything besides cpu, e.g. gpu, to fix the bugs in #742 #745 #687

@NickleDave
WIP: Add config/trainer.py with TrainerConfig
b7f43a9
@NickleDave
Rename common.device -> common.accelerator, return 'gpu' not 'cuda' i…
54b8718 
…f torch.cuda.is_available
@NickleDave
Fix config section in doc/api/index.rst
86b40c2
@NickleDave
Import trainer and TrainerConfig in src/vak/config/__init__.py, add t…
cf8f79c 
…o __all__
@NickleDave
Add pytorch-lightning to intersphinx in doc/conf.py
5b35753
@NickleDave
Fix cross-ref in docstring in src/vak/prep/frame_classification/make_…
8549cb4 
…splits.py: :constant: -> :const:
@NickleDave
Make lightning a dependency, instead of pytorch_lightning; import lig…
23b78a3 
…htning.pytorch everywhere instead of pytorch_lightning as lightning -- trying to make it so we can resolve API correctly in docstrings
@NickleDave
Fix in doc/api/index.rst: common.device -> common.accelerator
9d06234
@NickleDave
Finish writing TrainerConfig class
ad0cf3e
@NickleDave
Add tests for TrainerConfig class
9fd60d2
@NickleDave
Add trainer sub-table to all configs in tests/data_for_tests/configs
64ee588
@NickleDave
Add trainer sub-table to all configs in doc/toml
68efbc0
@NickleDave
Add trainer sub-table in config/valid-version-1.0.toml, rename -> val…
a6337f5 
…id-version-1.1.toml
@NickleDave
Remove device key from top-level tables in config/valid-version-1.1.toml
7094a6d
@NickleDave
Remove device key from top-level tables in tests/data_for_tests/configs
c14348e
@NickleDave
Remove 'device' key from configs in doc/toml
5d378e9
@NickleDave
Add 'trainer' attribute to EvalConfig, an instance of TrainerConfig; …
cfd1972 
…remove 'device' attribute
@NickleDave
Add 'trainer' attribute to PredictConfig, an instance of TrainerConfi…
96df71e 
…g; remove 'device' attribute
@NickleDave
Add 'trainer' attribute to TrainConfig, an instance of TrainerConfig;…
7c10814 
… remove 'device' attribute
@NickleDave
Fix typo in docstring in src/vak/config/train.py
fd12f97
@NickleDave
Add 'trainer' attribute to LearncurveConfig, an instance of TrainerCo…
7253c07 
…nfig; remove 'device' attribute. Also clean up docstring, removing attributes that no longer exist
@NickleDave
Remove device attribute from TrainConfig docstring
ef05f6f
@NickleDave
Fix VALID_TOML_PATH in config/validators.py -> 'valid-version-1.1.toml'
81b9a7d
@NickleDave
Fix how we instantiate TrainerConfig classes in from_config_dict meth…
5cd078d 
…od of EvalConfig/LearncurveConfig/PredictConfig/TrainConfig
@NickleDave
Fix typo in src/vak/config/valid-version-1.1.toml: predictor -> predict
436623a
@NickleDave
Fix unit tests after adding trainer attribute that is instance of Tra…
bd5f7b0 
…inerConfig
@NickleDave
Change src/vak/train/frame_classification.py to take trainer_config a…
2930eee 
…rgument
@NickleDave
Change src/vak/train/parametric_umap.py to take trainer_config argument
13a30ec
@NickleDave
Change src/vak/train/train_.py to take trainer_config argument
07031e1
@NickleDave
Fix src/vak/cli/train.py to pass trainer_config.asdict() into vak.tra…
f9400df 
…in.train_.train
@NickleDave
Replace 'device' with 'trainer_config' in vak/eval
5ac40e0
@NickleDave
Fix cli.eval to pass trainer_config into eval.eval_.eval
46ed723
@NickleDave
Replace 'device' with 'trainer_config' in vak/predict
994b69e
@NickleDave
Fix cli.predict to pass trainer_config into predict.predict_.predict
9a09ba1
@NickleDave
Replace 'device' with 'trainer_config' in vak/learncurve
a50eb34
@NickleDave
Fix cli.learncurve to pass trainer_config into learncurve.learncurve.…
64e9331 
…learning_curve
@NickleDave
Rename/replace 'device' fixture with 'trainer' fixture in tests/
22e9faa
@NickleDave
Use config.table.trainer attribute throughout tests, remove config.ta…
9a157f2 
…ble.device attribute that no longer exists
@NickleDave
Fix value for devices in fixtures/trainer.py: when device is 'cpu' tr…
8e96836 
…ainer must be > 0
@NickleDave
Fix default devices value for when accelerator is cpu in TrainerConfig
134dc0c
@NickleDave
Fix unit tests for TrainerConfig after fixing default devices for acc…
3c4ee20 
…elerator=cpu
@NickleDave
Fix default value for 'devices' set to -1 in some unit tests where we…
d92af7a 
… over-ride config in toml file
@NickleDave
fixup use config.table.trainer attribute throughout tests -- missed o…
ab12771 
…ne place in tests/test_eval/
@NickleDave
Add back 'device' fixture so we can use it to test Model class
51b888d
@NickleDave
Fix unit tests in test_models/test_base.by that literally used device…
5a243a9 
… to put tensors on device, not to change a config
@NickleDave
Fix assertion in tests/test_models/test_tweetynet.py, from where we s…
151abe2 
…witched to using lightning as the dependency
@NickleDave
Fix test for DiceLoss, change trainer_type fixture back to device fix…
316e7a3 
…ture
@NickleDave NickleDave merged commit e74b92b into main May 5, 2024
0 of 4 checks passed
This was referenced May 5, 2024
Closed
Closed
Closed
NickleDave added a commit that referenced this pull request May 6, 2024
@NickleDave
DOC: Update CHANGELOG after merging #752 [skip ci]
d34c3e7
@NickleDave NickleDave deleted the add-trainer-table-config branch August 24, 2024 12:52
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
1 participant
@NickleDave