EgoBench / analysis_scripts /calc_user_perf.py
emodiary's picture
Initial complete EgoBench release
3c4b9d3 verified
Raw History Blame Contribute Delete
1.75 kB
"""Print the four simulated-user averages saved by the main evaluator."""
import argparse
import json
import os
USER_PERFORMANCE_KEYS = (
"role_consistency",
"instruction_following",
"resilience",
"contextual_robustness",
)
def main():
parser = argparse.ArgumentParser(
description="Print simulated-user metric averages from evaluation summaries"
)
parser.add_argument(
"--eval_root",
default=os.path.join(os.path.dirname(__file__), "..", "eval_result"),
help="Directory containing per-model evaluation summaries",
)
args = parser.parse_args()
if not os.path.isdir(args.eval_root):
print(f"Evaluation directory not found: {args.eval_root}")
return
found = False
for model_name in sorted(os.listdir(args.eval_root)):
summary_path = os.path.join(args.eval_root, model_name, "summary.json")
if not os.path.isfile(summary_path):
continue
with open(summary_path, "r", encoding="utf-8") as stream:
summary = json.load(stream).get("summary", {})
averages = summary.get("avg_user_performance")
if not isinstance(averages, dict):
continue
counts = summary.get("user_performance_sample_counts", {})
found = True
print(f"\n{model_name}")
print(f"{'Metric':<28} {'Average':>10} {'Samples':>10}")
print("-" * 50)
for key in USER_PERFORMANCE_KEYS:
print(
f"{key:<28} {averages.get(key, 0.0):>10.4f} "
f"{counts.get(key, 0):>10}"
)
if not found:
print("No evaluation summary containing simulated-user metrics was found.")
if __name__ == "__main__":
main()