File size: 1,252 Bytes
8c10cf2 | 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 | # BookWriter-H_A2A (RQ1)
This folder contains analysis scripts and configuration used to evaluate BookWriter-H_A2A experiment outputs under the `RESULTS/` directory.
## Contents
- `reference_trajectory.yaml`
- Reference trajectory configuration used by the trajectory evaluator.
- `evaluate_trajectory-mix.py`
- Evaluates trajectory metrics across models.
- Outputs a CSV file (Markdown output is disabled).
- `evaluate_success.py`
- Computes success rate per model by checking generated chapter outputs.
- Outputs JSON and CSV summaries.
- `analyze_retry_patterns.py`
- Analyzes retry markers and error locations from `execution_path.md` files.
- Outputs JSON and CSV summaries.
## Typical Usage
- Trajectory evaluation:
- `python3 evaluate_trajectory-mix.py --config reference_trajectory.yaml --format csv --output evaluation_results.csv`
- Success rate analysis:
- `python3 evaluate_success.py`
- Retry pattern analysis:
- `python3 analyze_retry_patterns.py`
## Notes
- All scripts assume the experiment outputs live under:
- `/Users/wzr/TOSEM-2025/RESULTS/<MODEL>/BookWriter-H_A2A/test_results/`
- Some outputs (e.g., `execution_path.md`) are produced by the upstream experiment runner; these scripts only read them.
|