LR module for Scratch

This work accompanies the corresponding Scratch repository. Within this specific repository, the LR module - responsible for calculating the appropriate statistical data - is provided.

From the lrmodule python module, several public API methods are exposed:

• get_lr_system: load a trained LR system from disk from a given folder;
• get_reference_data: load reference data from disk;
• get_validation_experiment: return an Experiment that builds and validates a model.

Local development

1. Install all dependencies using pdm sync -G dev (to install dev dependencies as well)
2. Run checks with pdm run check or pdm check-quality to automatically fix the things as well
3. Run tests with pdm run test
4. To run everything pdm run all (no auto fixes) or with fixes: pdm run all-fix

All typing, linting and formatting configuration was taken from the Scratch repository for seamless integration.

Model validation

1. Update hyperparameters, experiment setup and data path as needed in models/[NAME]/validation.yaml;
2. Run validation experiments as pdm run lir models/[NAME]/validation.yaml;
3. Inspect the results in the output folder;
4. When satisfied, update the stored model (TODO).

Debugging in PyCharm

For example, instead of pdm run lir models/[NAME]/validation.yaml:

• Create a new 'Run/Debug configuration', with the following settings:
• Select the 'Python' template/default to start from
• Instead of 'script' select 'module', and specify lir
• As 'Script parameters', specify lrmodule/models/[NAME]/validation.yaml
• Make that the path to this repo is used as 'Working directory'

See the PyCharm documentation for more info.