File size: 4,668 Bytes
36ca9e1 | 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 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 | """
eval_postprocess.py
-------------------
Compute clip_rate_095, clip_rate_099, detach_rate, action_max_mean, and
palm_handle_dist_max_mean from one or more per-step metrics CSVs that
``evaluate_ppo_baseline.py`` writes via ``--log_csv``.
Usage:
python scripts/eval_postprocess.py CSV [CSV ...]
Prints a short JSON-per-CSV line and a Markdown summary table on stdout.
"""
import argparse
import csv
import json
import sys
from collections import defaultdict
from pathlib import Path
from statistics import mean
def load_rows(path: Path):
with open(path, newline="") as f:
reader = csv.DictReader(f)
return [
{
"episode": int(r["episode"]),
"step": int(r["step"]),
"reward": float(r["reward"]),
"done": int(r["done"]),
"action_abs_max": float(r["action_abs_max"]),
"palm_to_handle_dist": float(r["palm_to_handle_dist"]),
"success": int(r["success"]),
"normalized_progress": float(r["normalized_progress"]),
}
for r in reader
]
def episode_stats(rows):
"""Return per-episode dicts with terminal reward, success flag, max palm dist."""
by_ep = defaultdict(list)
for r in rows:
by_ep[r["episode"]].append(r)
eps = []
for ep, ep_rows in sorted(by_ep.items()):
ep_rows.sort(key=lambda r: r["step"])
last = ep_rows[-1]
# detach gives r_detach=-50; r_dist+r_act+r_time at slip step are tiny,
# so a step reward < -40 reliably flags detach. Min step reward is robust.
min_reward = min(r["reward"] for r in ep_rows)
any_success = any(r["success"] for r in ep_rows)
max_palm = max(r["palm_to_handle_dist"] for r in ep_rows)
eps.append({
"episode": ep,
"steps": len(ep_rows),
"terminal_reward": last["reward"],
"min_reward": min_reward,
"any_success": any_success,
"max_palm_dist": max_palm,
})
return eps
def summarize(path: Path):
rows = load_rows(path)
eps = episode_stats(rows)
n_steps = len(rows)
if n_steps == 0:
return {"path": str(path), "error": "empty"}
clip095 = sum(1 for r in rows if r["action_abs_max"] >= 0.95) / n_steps
clip099 = sum(1 for r in rows if r["action_abs_max"] >= 0.99) / n_steps
action_max_mean_step = mean(r["action_abs_max"] for r in rows)
n_eps = len(eps)
detach_rate = (
sum(1 for e in eps if (not e["any_success"]) and e["min_reward"] <= -40.0)
/ n_eps
)
success_rate = sum(1 for e in eps if e["any_success"]) / n_eps
palm_max_mean = mean(e["max_palm_dist"] for e in eps)
length_mean = mean(e["steps"] for e in eps)
return {
"path": str(path),
"episodes": n_eps,
"n_steps": n_steps,
"success_rate": success_rate,
"length_mean": length_mean,
"action_max_mean_step": action_max_mean_step,
"clip_rate_095": clip095,
"clip_rate_099": clip099,
"detach_rate": detach_rate,
"palm_handle_dist_max_mean": palm_max_mean,
}
def main():
p = argparse.ArgumentParser()
p.add_argument("csvs", nargs="+", type=Path)
p.add_argument("--summary_json", default=None, type=Path,
help="Optional: dump a list of summaries to this path.")
args = p.parse_args()
results = []
for path in args.csvs:
if not path.exists():
print(f"[skip] missing: {path}", file=sys.stderr)
continue
results.append(summarize(path))
print("\n| csv | eps | success | length | act_max | clip095 | clip099 | detach | palm_max |")
print("|---|---:|---:|---:|---:|---:|---:|---:|---:|")
for r in results:
if "error" in r:
print(f"| {Path(r['path']).name} | ERROR ({r['error']}) | | | | | | | |")
continue
print(
f"| {Path(r['path']).name} | "
f"{r['episodes']} | "
f"{r['success_rate']:.2f} | "
f"{r['length_mean']:.1f} | "
f"{r['action_max_mean_step']:.3f} | "
f"{r['clip_rate_095']:.3f} | "
f"{r['clip_rate_099']:.3f} | "
f"{r['detach_rate']:.2f} | "
f"{r['palm_handle_dist_max_mean']:.3f} |"
)
print("\nJSON:")
for r in results:
print(json.dumps(r))
if args.summary_json:
args.summary_json.parent.mkdir(parents=True, exist_ok=True)
with open(args.summary_json, "w") as f:
json.dump(results, f, indent=2)
if __name__ == "__main__":
main()
|