Datasets:
| """ | |
| Minimal example: compute accuracy on ChartReal using normalize_answer. | |
| Usage: | |
| python3 eval/eval_example.py path/to/predictions.jsonl path/to/data.jsonl | |
| """ | |
| import json | |
| import sys | |
| from normalize import normalize_answer | |
| def main(): | |
| if len(sys.argv) != 3: | |
| print("Usage: python3 eval_example.py <predictions.jsonl> <data.jsonl>") | |
| sys.exit(1) | |
| preds_path, data_path = sys.argv[1], sys.argv[2] | |
| # Index gold answers by example_id | |
| gold = {} | |
| with open(data_path, encoding="utf-8") as f: | |
| for line in f: | |
| d = json.loads(line) | |
| gold[d["example_id"]] = d | |
| # Score predictions | |
| n = c = 0 | |
| with open(preds_path, encoding="utf-8") as f: | |
| for line in f: | |
| p = json.loads(line) | |
| ex = gold.get(p["example_id"]) | |
| if not ex: | |
| continue | |
| n += 1 | |
| if normalize_answer( | |
| p["prediction_raw"], ex["answer"], ex["answer_type"], | |
| gold_numeric=ex.get("answer_numeric"), | |
| ): | |
| c += 1 | |
| print(f"Accuracy: {c}/{n} = {c/n*100:.2f}%") | |
| if __name__ == "__main__": | |
| main() | |