Research – Pilot plan: alternative fitness metrics for SR

[parisazeynaly]

Goal: Compare MSE vs correlation/similarity metrics (Pearson, Spearman, Cosine; optional Distance Corr) for symbolic regression, and observe effects on search dynamics.

Plan:

• Implement metrics in src/gp/metrics.py + unit tests.
• Pilot on 2 SR datasets (Nguyen-1, Feynman-III-1), seed=0, budget≈200 evals.
• Log per-generation CSV + short summary (table/plot).
• Expose metric choice via config/CLI.

Questions:

1. Preferred SR datasets/seeds for the first run?
2. Config vs CLI (or both) for metric selection?
3. Any existing evaluation hook you prefer I extend?

Status: env ready; onboarding PR submitted (#30). Waiting for your guidance before coding.