|
|
|
|
|
|
| import json
|
| from pathlib import Path
|
| import numpy as np
|
| import matplotlib.pyplot as plt
|
|
|
| MONTHS = ['2407','2408','2409','2410','2411','2412',
|
| '2501','2502','2503','2504','2505','2506']
|
|
|
| def load_values_jsons(root: Path):
|
| """扫描 root 下所有子目录中的 values.json"""
|
| subfolders = sorted([p for p in root.iterdir() if p.is_dir()])
|
| all_data = {}
|
| for folder in subfolders:
|
| vpath = folder / "values.json"
|
| if not vpath.exists():
|
| continue
|
| try:
|
| data = json.loads(vpath.read_text(encoding="utf-8"))
|
| except Exception as e:
|
| print(f"[WARN] 无法解析 {vpath}: {e}")
|
| continue
|
| all_data[folder.name] = data
|
| return all_data
|
|
|
|
|
| def build_matrix(all_data, metric="acc_norm", prefix="arxiv_mc_"):
|
| """
|
| 构造矩阵：
|
| 行：子文件夹名（不同模型或实验）
|
| 列：固定的12个月份
|
| """
|
| folder_names = list(all_data.keys())
|
| n_rows = len(folder_names)
|
| n_cols = len(MONTHS)
|
| mat = np.full((n_rows, n_cols), np.nan)
|
|
|
| for i, folder in enumerate(folder_names):
|
| data = all_data[folder]
|
| items = data.get("tasks", [])
|
| d = {(it.get("task"), it.get("metric")): it.get("value")
|
| for it in items if it.get("task")}
|
| for j, month in enumerate(MONTHS):
|
| task_name = f"{prefix}{month}"
|
| val = d.get((task_name, metric))
|
| if val is not None:
|
| mat[i, j] = float(val)
|
| return folder_names, mat
|
|
|
|
|
| def plot_heatmap(mat, row_labels, col_labels, out_png, title="Accuracy heatmap"):
|
| fig, ax = plt.subplots(figsize=(max(10, len(col_labels)*0.7), max(6, len(row_labels)*0.4)))
|
| im = ax.imshow(mat, aspect="auto", cmap="viridis")
|
|
|
| ax.set_xticks(range(len(col_labels)))
|
| ax.set_yticks(range(len(row_labels)))
|
| ax.set_xticklabels(col_labels, rotation=45, ha="right")
|
| ax.set_yticklabels(row_labels)
|
| ax.set_xlabel("Eval month")
|
| ax.set_ylabel("Experiment / checkpoint")
|
| ax.set_title(title)
|
|
|
|
|
| for i in range(mat.shape[0]):
|
| for j in range(mat.shape[1]):
|
| val = mat[i, j]
|
| if not np.isnan(val):
|
| ax.text(j, i, f"{val:.2f}", ha="center", va="center",
|
| fontsize=7, color="white")
|
|
|
| fig.colorbar(im, ax=ax, fraction=0.046, pad=0.04, label="value")
|
| fig.tight_layout()
|
| fig.savefig(out_png, dpi=200)
|
| plt.close(fig)
|
| print(f"✅ Saved heatmap to {out_png}")
|
|
|
|
|
|
|
|
|
| def main():
|
| root = Path("/mnt/data/litgpt/out/eval/qwen2_7b_question_focus/")
|
| all_data = load_values_jsons(root)
|
| folders, mat = build_matrix(all_data, metric="acc", prefix="arxiv_mc_")
|
| plot_heatmap(mat, folders, MONTHS, root / "acc_heatmap.png",
|
| title="Accuracy (acc) 2407–2506")
|
| folders, mat = build_matrix(all_data, metric="acc_norm", prefix="arxiv_mc_")
|
| plot_heatmap(mat, folders, MONTHS, root / "acc_norm_heatmap.png",
|
| title="Accuracy (acc_norm) 2407–2506")
|
|
|
|
|
|
|
| if __name__ == "__main__":
|
| main()
|
|
|
|
|
