Add EMNISTDataModule

[sugatoray]

What does this PR do?

Closes #672, #676 and #685.

A summary of changes and modifications ⭐ 🔥 [CLICK TO EXPAND]

• File Added:

  • pl_bolts/datasets/emnist_dataset.py 🟢
  • Contents:
    • EMNIST_METADATA
    • EMNIST dataset
    • BinaryEMNIST dataset Need New PR or add to #672 ⚠️

• File Added:

  • pl_bolts/datamodules/emnist_dataset.py 🟢
  • Contents:
    • EMNISTDataModule
    • BinaryEMNISTDataModule Need New PR or add to #672 ⚠️

• Files Modified

  • Package: pl_bolts

    • pl_bolts/datasets/__init__.py 🟢
    • pl_bolts/datamodules/__init__.py 🟢

  • Tests:

    • For datamodules:
      • tests/datamodules/test_imports.py 🟢
      • tests/datamodules/test_datamodules.py WIP 🟠

Adding BinaryEMNIST and BinaryEMNISTDataModule was logical, looking at how MNIST and BinaryMNIST (dataset and datamodules) were implemented.

About the dataset

source: https://arxiv.org/pdf/1702.05373.pdf [Table-I]

source: https://arxiv.org/pdf/1702.05373.pdf [Table-II]

Before submitting

• Was this discussed/approved via a Github issue? (no need for typos and docs improvements) Make EMNISTDataModule #672
• Did you read the contributor guideline, Pull Request section? Y 🟢
• Did you make sure your PR does only one thing, instead of bundling different changes together? Y 🟢
• Did you make sure to update the documentation with your changes? Y 🟢
• Did you write any new necessary tests? [not needed for typos/docs] Y 🟢
• Did you verify new and existing tests pass locally with your changes? Y 🟢
• If you made a notable change (that affects users), did you update the CHANGELOG? Y 🟢

PR review

• Is this pull request ready for review? (if not, please submit in draft mode) READY 🟢

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

[pep8speaks]

Hello @sugatoray! Thanks for updating this PR.

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2021-08-13 13:04:38 UTC

[codecov]

Codecov Report

Merging #676 (ef13456) into master (bd28835) will increase coverage by 47.35%.
The diff coverage is 80.00%.

❗ Current head ef13456 differs from pull request most recent head 5833e60. Consider uploading reports for the commit 5833e60 to get more accurate results

@@             Coverage Diff             @@
##           master     #676       +/-   ##
===========================================
+ Coverage   24.32%   71.67%   +47.35%     
===========================================
  Files         120      120               
  Lines        7486     7397       -89     
===========================================
+ Hits         1821     5302     +3481     
+ Misses       5665     2095     -3570     

Flag	Coverage Δ
cpu	71.67% <80.00%> (+47.35%)	⬆️
pytest	71.67% <80.00%> (+47.35%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
pl_bolts/transforms/dataset_normalizations.py	72.00% <20.00%> (+37.00%)	⬆️
pl_bolts/datasets/emnist_dataset.py	90.90% <90.90%> (ø)
pl_bolts/datamodules/__init__.py	100.00% <100.00%> (ø)
pl_bolts/datasets/__init__.py	100.00% <100.00%> (ø)
pl_bolts/models/rl/advantage_actor_critic_model.py
pl_bolts/losses/self_supervised_learning.py	71.33% <0.00%> (+0.18%)	⬆️
pl_bolts/models/self_supervised/cpc/cpc_module.py	20.58% <0.00%> (+0.58%)	⬆️
...l_bolts/models/self_supervised/byol/byol_module.py	22.34% <0.00%> (+0.83%)	⬆️
pl_bolts/models/vision/image_gpt/igpt_module.py	20.65% <0.00%> (+0.87%)	⬆️
... and 70 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update bd28835...5833e60. Read the comment docs.

[akihironitta]

LGTM!

Description of the change: Introduced a new arg strict_val_split: bool to use the balanced validation set defined in the EMNIST paper.

(always open for any discussions/comments on the API (strict_val_split) :])

---

Note: The failing tests are irrelevant to the changes in this PR.

[sugatoray]

@akihironitta Looks good to me. Thank you, for working on it. I am not sure, if you were waiting for any feedback from me. It seems that this has not been reviewed by other reviewers since your approval. Please let me know if there is something I need to do.

[akihironitta]

@sugatoray Hi, thanks for having a look again!

@PyTorchLightning/core-bolts Would you mind having a look at the changes?

[akihironitta]

The failing tests are irrelevant to the change in this PR.

pytest summary at c6adbae

=========================== short test summary info ============================
FAILED tests/callbacks/verification/test_base.py::test_verification_base_get_input_array[device0]
FAILED tests/callbacks/verification/test_batch_gradient.py::test_batch_gradient_verification_pl_module[device0-True]
FAILED tests/callbacks/verification/test_batch_gradient.py::test_batch_gradient_verification_pl_module[device0-False]
FAILED tests/callbacks/verification/test_batch_gradient.py::test_batch_gradient_verification_callback[0]
FAILED tests/models/rl/test_scripts.py::test_cli_run_rl_dqn[ --env PongNoFrameskip-v4 --max_steps 10 --fast_dev_run 1 --warm_start_size 10 --n_steps 2 --batch_size 10]
FAILED tests/models/rl/test_scripts.py::test_cli_run_rl_reinforce[ --env CartPole-v0 --max_steps 10 --fast_dev_run 1 --batch_size 10]
FAILED tests/models/rl/test_scripts.py::test_cli_run_rl_vanilla_policy_gradient[ --env CartPole-v0 --max_steps 10 --fast_dev_run 1 --batch_size 10]
= 7 failed, 288 passed, 23 skipped, 1 xfailed, 532 warnings in 1053.39s (0:17:33) =

[akihironitta]

@PyTorchLightning/core-bolts Could you have another look?