Spaces:
Sleeping
Sleeping
File size: 2,588 Bytes
a6f0611 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 | """Plot training curves from runs/metrics.jsonl.
Outputs:
runs/reward_curve.png - reward per episode + rolling mean
runs/loss_curve.png - 1.0 - grader_score (proxy "loss") per episode
Usage:
python plot_metrics.py [--input runs/metrics.jsonl] [--out runs/]
"""
from __future__ import annotations
import argparse
import json
from pathlib import Path
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
def _rolling(xs, w):
out, acc = [], []
for x in xs:
acc.append(x)
if len(acc) > w:
acc.pop(0)
out.append(sum(acc) / len(acc))
return out
def main() -> None:
p = argparse.ArgumentParser()
p.add_argument("--input", default="runs/metrics.jsonl")
p.add_argument("--out", default="runs")
p.add_argument("--window", type=int, default=10)
args = p.parse_args()
src = Path(args.input)
out = Path(args.out)
out.mkdir(parents=True, exist_ok=True)
rows = [json.loads(l) for l in src.read_text().splitlines() if l.strip()]
if not rows:
raise SystemExit(f"No rows in {src}")
eps = [r.get("episode", i + 1) for i, r in enumerate(rows)]
rewards = [float(r.get("reward", 0.0)) for r in rows]
plt.figure(figsize=(8, 4.5))
plt.plot(eps, rewards, alpha=0.4, label="reward")
plt.plot(eps, _rolling(rewards, args.window), linewidth=2, label=f"rolling mean (w={args.window})")
plt.axhline(0, color="gray", linewidth=0.5)
plt.xlabel("episode"); plt.ylabel("reward"); plt.title("Training reward")
plt.legend()
plt.savefig(out / "reward_curve.png", dpi=120)
plt.close()
# Proxy loss = 1 - normalized grader score; if missing, derive from recall+precision
losses = []
for r in rows:
g = r.get("grader_score")
if g is None:
recall = float(r.get("recall", 0.0))
precision = float(r.get("precision", 0.0))
g = 0.55 + 0.20 * recall + 0.15 * precision if (recall >= 0.8 and precision >= 0.7) else 0.30 * recall + 0.10 * precision
losses.append(max(0.0, 1.0 - float(g)))
plt.figure(figsize=(8, 4.5))
plt.plot(eps, losses, alpha=0.4, label="loss (1 - grader)")
plt.plot(eps, _rolling(losses, args.window), linewidth=2, label=f"rolling mean (w={args.window})")
plt.xlabel("episode"); plt.ylabel("loss"); plt.title("Training loss proxy")
plt.legend()
plt.savefig(out / "loss_curve.png", dpi=120)
plt.close()
print(f"wrote {out / 'reward_curve.png'}")
print(f"wrote {out / 'loss_curve.png'}")
if __name__ == "__main__":
main()
|