def create_metrics_table()

in project/paperbench/experiments/judge_eval/judge_eval_perf_cost.py [0:0]

• 22 lines of code
• 3 McCabe index (conditional complexity)

def create_metrics_table(model_results):
    """
    Creates a DataFrame with performance and cost metrics for each model
    """
    data = []
    # Use the same model order as in the plot
    for model in MODELS_SORTED:
        stats = model_results[model]
        model_name = stats["model_name"]
        reasoning_effort = stats["reasoning_effort"]
        f1_score = stats["aggregate_metrics"]["f1"]
        cost = compute_cost(stats["aggregate_token_usage"]["mean"], MODEL_COST_PER_TOKEN)

        # Use the same model name mapping as in the plot
        display_name = MODEL_NAME_TO_LABEL[model_name]

        data.append(
            {
                "Model": display_name,
                "Reasoning Effort": reasoning_effort if reasoning_effort else "default",
                "F1 Score": f1_score,
                "Avg Cost Per Paper (USD)": cost,
            }
        )

    df = pd.DataFrame(data)

    # No need to sort since we're using MODELS_SORTED order

    # Save to CSV
    output_path = "experiments/judge_eval/tables/perf_cost_table.csv"
    df.to_csv(output_path, index=False, float_format="%.4f")
    print(f"Saved metrics table to '{output_path}'")

    return df