Main

[maxiallard]

This pull request introduces improvements to logging, expands support for sparse matrices in the scgpt dataset, and adds a new test for quantization initialization. The most impactful changes are the enhanced handling of sparse matrices for efficient data access and serialization, and the refinement of logging configuration for better clarity and control. Additionally, the package version is bumped and a dependency is added.

Logging improvements:

• Refined logging setup in helical/__init__.py to use a dedicated helical logger, improved formatting to include timestamps, and disabled propagation to the root logger for clearer, project-specific log output.
• Updated logging usage in helical/models/transcriptformer/utils/utils.py to use a module-specific logger instead of global configuration.

Sparse matrix support and dataset enhancements:

• Added efficient handling of sparse matrices in scgpt dataset (helical/models/scgpt/dataset.py): tracks matrix type, handles serialization/deserialization, and retrieves rows as dense arrays only when needed, improving memory usage and performance. [1] [2]
• Adjusted process_data in scgpt/model.py to use the raw adata.X matrix, supporting both sparse and dense formats.

Testing improvements:

• Added a test for model initialization with quantization in test_c2s.py, verifying correct device assignment and quantization configuration.

Package management:

• Bumped package version to 1.4.15 and added protobuf>=3.20.0 to dependencies in pyproject.toml. [1] [2]