Tune augmentations with CLI and config for contrastive models

[ziw-liu]

A simple CLI is added for training contrastive models with triplet data (not accessible from the main viscy entry point and not working for all stages). This allows hyperparameters to be documented in human-readable YAML config files along side TensorBoard training logs.

Usage:

python -m viscy.cli.contrastive_triplet fit -c fit.yml

Example training run:

/hpc/projects/intracellular_dashboard/viral-sensor/infection_classification/models/contrastive_tune_augmentations

For the future iteration (v0.3?) it might make sense to replace the manual CLI entry point (viscy [command]) with python's built-in (python -m viscy.cli.module [command]) to make supporting multiple learning tasks easier.

[mattersoflight]

@ziw-liu please version an example fit.yml, perhaps in the examples folder.

[Soorya19Pradeep]

I used the config provided on the path and tried the cli with my viscy environment. I tried a few changes in augmentation and configuring the workers. Thanks @ziw-liu !

[ziw-liu]

@ziw-liu please version an example fit.yml, perhaps in the examples folder.

The config is in the linked training run directory.