Add support for mouse triplets and mabe22

[gchindemi]

This pull request introduces support for the MABe22 dataset and refines data preprocessing, scaling, and training configurations. Key changes include adding a new dataset, enhancing scaling logic, updating the input pipeline for feature selection, and improving test coverage.

New Dataset Support:

• Added support for the MABe22 dataset, including its preprocessing and loading logic in a new module src/lisbet/datasets/mabe22.py. This includes methods to handle training and testing data, along with metadata and references.
• Updated fetch_dataset in src/lisbet/datasets/core.py to include MABe22 dataset retrieval and preprocessing logic.
• Added MABe22 to the dataset options in the CLI (src/lisbet/cli.py).

Data Preprocessing and Scaling:

• Enhanced _load_posetracks in src/lisbet/datasets/core.py to support scaling based on explicit factors or the image_size_px attribute. This ensures more flexible and accurate coordinate normalization.
• Added image_size_px attribute to CalMS21 preprocessing to enable consistent scaling.

Input Pipeline Refinements:

• Updated __init__ in src/lisbet/input_pipeline.py to compute and store individual-specific feature indices for efficient feature selection.
• Modified __getitem__ in src/lisbet/input_pipeline.py to apply swaps for specific individuals' features, improving clarity and precision.

Training Configuration:

• Added a pin_memory parameter to _configure_dataloaders in src/lisbet/training.py, dynamically set based on the device type (e.g., CUDA, MPS, or CPU).

Test Enhancements:

• Added tests to validate scaling logic, ensuring explicit scaling and image_size_px scaling produce consistent results and raise errors for invalid data.
• Improved error message validation in inference tests.

[Copilot]

Pull Request Overview

Add support for the MABe22 Mouse Triplets dataset, refine scaling and preprocessing logic, and update training and input pipeline configurations.

• Introduce a new mabe22 module with loading and preprocessing for mouse triplets and register it in the CLI.
• Enhance _load_posetracks to handle explicit scaling, image_size_px scaling, and auto-scaling with consistent error checks.
• Compute per-individual feature indices in BaseDataset and update swap/delay datasets to target specific individual features; add dynamic pin_memory selection for dataloaders.

Reviewed Changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
tests/test_training_integration.py	Updated sample dataset path in integration test to include the datasets/ prefix
tests/test_training_helpers.py	Updated _configure_dataloaders call to include the new pin_memory argument
tests/test_input_pipeline.py	Added tests for multi-individual feature indexing, swap-only and delay-only behaviors
tests/test_inference.py	Adjusted error message regex to remove the exclamation mark
tests/test_datasets_core.py	Added tests to validate explicit vs. image_size_px scaling and out-of-range checks
src/lisbet/training.py	Added pin_memory parameter to _configure_dataloaders and set it dynamically based on device type
src/lisbet/input_pipeline.py	Compute per-individual feature indices in BaseDataset; update swap/delay classes to swap by index
src/lisbet/datasets/mabe22.py	New module implementing support for the MABe22 dataset
src/lisbet/datasets/core.py	Enhanced coordinate scaling logic and integrated MABe22 fetch path
src/lisbet/datasets/calms21.py	Added image_size_px attribute to CalMS21 preprocessing
src/lisbet/cli.py	Registered MABe22_MouseTriplets dataset in the CLI

[Copilot]

Pull Request Overview

This PR adds support for the new MABe22 dataset, updates data preprocessing (including scaling logic), refines the input pipeline, and improves test coverage for training, inference, and evaluation. Key changes include adding a new module for the MABe22 dataset, updating scaling/normalization routines in dataset loaders, and modifying the data loader configuration with a new pin_memory parameter.

Reviewed Changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
tests/test_training_integration.py	Updated test path to reflect new directory structure for sample data
tests/test_training_helpers.py	Updated dataloader call to include pin_memory parameter
tests/test_inference.py	Adjusted expected error message for inference failure
tests/test_input_pipeline.py	Added tests verifying individual indices and swap/shift behavior
tests/test_datasets_core.py	Added tests for explicit scaling behavior and equivalence with image_size_px scaling
src/lisbet/training.py	Updated dataloader configuration and pin_memory assignment based on device type
src/lisbet/input_pipeline.py	Revised swapping/shifting to directly target the second individual’s features
src/lisbet/evaluation.py	Updated error label extraction to match new annotation naming convention
src/lisbet/datasets/mabe22.py	New module implementing data processing for the MABe22 dataset
src/lisbet/datasets/core.py	Extended scaling logic to handle explicit scale strings and image_size_px attributes
src/lisbet/datasets/calms21.py	Updated preprocessing to include image_size_px and new target_cls naming
src/lisbet/cli.py	Updated dataset options to add support for MABe22_MouseTriplets

Comments suppressed due to low confidence (1)

tests/test_inference.py:127

• The updated error message no longer includes an exclamation mark. Please verify that this change is consistent with the intended exception messaging throughout the codebase.

with pytest.raises(ValueError, match="Incompatible input features"):