[CPBench] Benchmarking Wrapper

[lehendo]

Benchmarking wrapper for the CPBench project.

Usage:

List what's available:

python benchmarks/cpbench/benchmark.py --list

Full benchmark run (trains from scratch, tests at α = 0.01, 0.05, 0.10, 0.20):

python benchmarks/cpbench/benchmark.py \
    --task sleep_staging_isruc \
    --method label \
    --data-path /path/to/ISRUC-I

Skip training with a saved checkpoint:

python benchmarks/cpbench/benchmark.py \
    --task sleep_staging_isruc \
    --method label \
    --data-path /path/to/ISRUC-I \
    --checkpoint /path/to/model.pth

Quick smoke-test (uses a tiny data subset, runs only α = 0.10):

python benchmarks/cpbench/benchmark.py \
    --task sleep_staging_isruc \
    --method base \
    --data-path /path/to/ISRUC-I \
    --dev

Save results to JSON for aggregating across runs:

python benchmarks/cpbench/benchmark.py \
    --task sleep_staging_isruc \
    --method label \
    --data-path /path/to/ISRUC-I \
    --output results/isruc_label.json