Re-evaluate breaking-IID experiments from checkpoints

[mcw92]

Additional metrics for breaking-IID series

As we decided to analyze a variety of performance metrics in addition to the plain accuracy, we need to re-evaluate all 16-node breaking-IID experiments from their respective model checkpoints to obtain the local and global confusion matrices.

Relevant scripts are:

• Actual Python script to run: ./scripts/examples/evaluate_from_checkpoint_breaking_iid.py
• SLURM utility function to obtain correct checkpoint paths for each parameter combination, starting from given base path: find_checkpoint_dir_and_uuid in ./specialcouscous/utils/slurm.py
• Job script generation script: ./scripts/experiments/generate_parallel_evaluation_from_breaking_iid_ckpt_job_scripts.py
• Path to actual results on HoreKa: ${BASEDIR}/results/ (BASEDIR corresponds to our workspace)