[siliataider]

This Pull request:

Changes:

Moves ml_dataloader.py from unittest to pytest to make the tests shorter a,nd easier to read and to extend.
The test logic is unchanged.

• Much shorter file: 8127 --> 4969 lines (-39%), with 104 tests in 74 test functions instead of 105
• Fixtures for the input data: each test can directly ask for a common input built with the fixtures instead of writing its own
• No more file cleanup code: each test runs in its own tmp_path, created and deleted by pytest
  • This removes all the try/except/finally blocks and the os.removes that existed only to delete temporary files
  • Tests can't clash on file names anymore
• One test class for both lazy and eager loading: the copy-pasted lazy/eager classes are merged into TestDataLoaderSingleDataframe and TestDataLoaderMultipleDataframes, which runn all of their tests twice (for both loading modes) using a parametrized fixture
• Skip instead of fail if package is missin: the PyTorch/TensorFlow/JAX tests are skipped when the framework isn't installed

Checklist:

• tested changes locally

[github-actions]

Test Results

    24 files      24 suites   3d 22h 30m 29s ⏱️
 3 903 tests  3 903 ✅ 0 💤 0 ❌
83 563 runs  83 563 ✅ 0 💤 0 ❌

Results for commit f98b9b8.

♻️ This comment has been updated with latest results.