"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"tab = pd.crosstab(pd.Series(groups, name=\"subject\"), pd.Series(np.array(CLASSES)[y], name=\"class\"))[CLASSES]\n",
"print(\"beats per subject: median\", int(tab.sum(axis=1).median()),\n",
" \"| min\", int(tab.sum(axis=1).min()), \"| max\", int(tab.sum(axis=1).max()))\n",
"for c in [\"S\", \"V\", \"F\"]:\n",
" top = tab[c].sort_values(ascending=False)\n",
" print(f\"{c}: {int((top > 0).sum())} of 43 subjects have any; \"\n",
" f\"top-2 subjects ({', '.join(top.index[:2])}) hold {100 * top.iloc[:2].sum() / top.sum():.0f}%\")\n",
"\n",
"fig, axes = plt.subplots(1, 4, figsize=(12, 2.8), sharey=True)\n",
"t_ms = (np.arange(X.shape[1]) * STEP - PRE) / 360 * 1000\n",
"for k, ax in enumerate(axes):\n",
" beats = X[y == k, :, 0]\n",
" ax.plot(t_ms, beats[:: max(1, len(beats) // 40)].T, color=\"0.8\", lw=0.5) # ~40 example beats\n",
" ax.plot(t_ms, beats.mean(axis=0), color=\"#F4511E\", lw=2)\n",
" ax.set_title(f\"{CLASSES[k]} (n={len(beats):,})\")\n",
" ax.set_xlabel(\"ms from R peak\")\n",
"axes[0].set_ylabel(\"standardised amplitude\")\n",
"plt.tight_layout()\n",
"plt.show()"
]
},
{
"cell_type": "markdown",
"id": "4d98dd76",
"metadata": {},
"source": [
"## 4. 兩種切法\n",
"\n",
"兩種切法都分成 3 折（每次 2/3 訓練、1/3 測試），差別只在「打散的單位」：\n",
"\n",
"- **切法 A：隨機切心跳**（`StratifiedKFold`）：所有心跳打散，各類別比例在每折相同。同一個人的心跳會同時出現在訓練與測試。\n",
"- **切法 B：依受試者切**（`StratifiedGroupKFold`，`groups` = 受試者）：同一個人的所有心跳只會在同一折。它是第 11 章分組 K-fold 的變形，會盡量讓各折的類別比例接近。\n",
"\n",
"先確認兩種切法下，測試集裡的人有沒有在訓練集出現過："
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "7168cf2f",
"metadata": {
"execution": {
"iopub.execute_input": "2026-10-06T15:12:38.567085Z",
"iopub.status.busy": "2026-10-06T15:12:38.566876Z",
"iopub.status.idle": "2026-10-06T15:12:38.986904Z",
"shell.execute_reply": "2026-10-06T15:12:38.986251Z"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"A: random beats fold 1: test beats 33,560 from 43 subjects; subjects also in training: 43; test class counts [30030 927 2336 267]\n",
"A: random beats fold 2: test beats 33,560 from 43 subjects; subjects also in training: 43; test class counts [30030 927 2336 267]\n",
"A: random beats fold 3: test beats 33,560 from 43 subjects; subjects also in training: 43; test class counts [30029 927 2336 268]\n",
"B: by subject fold 1: test beats 33,421 from 15 subjects; subjects also in training: 0; test class counts [29498 1463 2072 388]\n",
"B: by subject fold 2: test beats 36,285 from 15 subjects; subjects also in training: 0; test class counts [33158 853 2247 27]\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"B: by subject fold 3: test beats 30,974 from 13 subjects; subjects also in training: 0; test class counts [27433 465 2689 387]\n"
]
}
],
"source": [
"splits = {\n",
" \"A: random beats\": list(StratifiedKFold(3, shuffle=True, random_state=RS).split(X, y)),\n",
" \"B: by subject\": list(StratifiedGroupKFold(3, shuffle=True, random_state=RS).split(X, y, groups)),\n",
"}\n",
"for name, folds in splits.items():\n",
" for i, (tr, te) in enumerate(folds):\n",
" assert len(set(tr) & set(te)) == 0 and len(tr) + len(te) == len(y)\n",
" shared = set(groups[tr]) & set(groups[te])\n",
" if name.startswith(\"B\"):\n",
" assert not shared # no subject on both sides\n",
" print(f\"{name} fold {i + 1}: test beats {len(te):,} from {len(set(groups[te]))} subjects; \"\n",
" f\"subjects also in training: {len(shared)}; test class counts {np.bincount(y[te], minlength=4)}\")"
]
},
{
"cell_type": "markdown",
"id": "f3b24d88",
"metadata": {},
"source": [
"切法 A 的每一折，測試集的 43 個人**全部**也在訓練集裡；切法 B 則沒有任何一個人重疊。也注意切法 B 各折的類別數差很多：F 類主要集中在少數人身上，某一折可能只分到幾十個。"
]
},
{
"cell_type": "markdown",
"id": "a778ac68",
"metadata": {},
"source": [
"## 5. 一維卷積網路（1D-CNN）\n",
"\n",
"和第 12 章的 CNN 幾乎一樣，只是濾鏡從 3×3 的方格變成**沿時間軸滑動的一小段**（`Conv1D`）。三層卷積的濾鏡長度分別是 7、5、5 個點（180 Hz 下約 39、28、28 毫秒），`GlobalAveragePooling1D` 把每個濾鏡在整個心跳上的反應取平均，最後 `Dense(4, softmax)` 輸出四類的機率。\n",
"\n",
"**處理不平衡**：在**切好之後**，只對訓練集的 N 類隨機抽 8,000 個（其他類全留），測試集完全不動，保持真實比例。若先抽樣再切，等於讓抽樣決定了誰當考題，也是一種洩漏的來源。"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "d50a7d64",
"metadata": {
"execution": {
"iopub.execute_input": "2026-10-06T15:12:38.989462Z",
"iopub.status.busy": "2026-10-06T15:12:38.989275Z",
"iopub.status.idle": "2026-10-06T15:12:39.067019Z",
"shell.execute_reply": "2026-10-06T15:12:39.066248Z"
}
},
"outputs": [
{
"data": {
"text/html": [
"Model: \"sequential\"\n",
"\n"
],
"text/plain": [
"\u001b[1mModel: \"sequential\"\u001b[0m\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
"┃ Layer (type) ┃ Output Shape ┃ Param # ┃\n",
"┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
"│ conv1d (Conv1D) │ (None, 117, 16) │ 128 │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ max_pooling1d (MaxPooling1D) │ (None, 58, 16) │ 0 │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ conv1d_1 (Conv1D) │ (None, 58, 32) │ 2,592 │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ max_pooling1d_1 (MaxPooling1D) │ (None, 29, 32) │ 0 │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ conv1d_2 (Conv1D) │ (None, 29, 32) │ 5,152 │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ global_average_pooling1d │ (None, 32) │ 0 │\n",
"│ (GlobalAveragePooling1D) │ │ │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ dense (Dense) │ (None, 4) │ 132 │\n",
"└─────────────────────────────────┴────────────────────────┴───────────────┘\n",
"\n"
],
"text/plain": [
"┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
"┃\u001b[1m \u001b[0m\u001b[1mLayer (type) \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
"┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
"│ conv1d (\u001b[38;5;33mConv1D\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m117\u001b[0m, \u001b[38;5;34m16\u001b[0m) │ \u001b[38;5;34m128\u001b[0m │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ max_pooling1d (\u001b[38;5;33mMaxPooling1D\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m58\u001b[0m, \u001b[38;5;34m16\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ conv1d_1 (\u001b[38;5;33mConv1D\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m58\u001b[0m, \u001b[38;5;34m32\u001b[0m) │ \u001b[38;5;34m2,592\u001b[0m │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ max_pooling1d_1 (\u001b[38;5;33mMaxPooling1D\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m29\u001b[0m, \u001b[38;5;34m32\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ conv1d_2 (\u001b[38;5;33mConv1D\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m29\u001b[0m, \u001b[38;5;34m32\u001b[0m) │ \u001b[38;5;34m5,152\u001b[0m │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ global_average_pooling1d │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m32\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │\n",
"│ (\u001b[38;5;33mGlobalAveragePooling1D\u001b[0m) │ │ │\n",
"├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
"│ dense (\u001b[38;5;33mDense\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m4\u001b[0m) │ \u001b[38;5;34m132\u001b[0m │\n",
"└─────────────────────────────────┴────────────────────────┴───────────────┘\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
" Total params: 8,004 (31.27 KB)\n",
"\n"
],
"text/plain": [
"\u001b[1m Total params: \u001b[0m\u001b[38;5;34m8,004\u001b[0m (31.27 KB)\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
" Trainable params: 8,004 (31.27 KB)\n",
"\n"
],
"text/plain": [
"\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m8,004\u001b[0m (31.27 KB)\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
" Non-trainable params: 0 (0.00 B)\n",
"\n"
],
"text/plain": [
"\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"N_KEEP = 8000 # normal beats kept in each training set (after splitting)\n",
"EPOCHS = 12\n",
"\n",
"\n",
"def build_model():\n",
" return keras.Sequential([\n",
" keras.Input(shape=(X.shape[1], 1)),\n",
" keras.layers.Conv1D(16, 7, padding=\"same\", activation=\"relu\"), # 16 filters, 7 points long\n",
" keras.layers.MaxPooling1D(2),\n",
" keras.layers.Conv1D(32, 5, padding=\"same\", activation=\"relu\"),\n",
" keras.layers.MaxPooling1D(2),\n",
" keras.layers.Conv1D(32, 5, padding=\"same\", activation=\"relu\"),\n",
" keras.layers.GlobalAveragePooling1D(), # average over time\n",
" keras.layers.Dense(4, activation=\"softmax\"),\n",
" ])\n",
"\n",
"\n",
"def run_fold(tr, te, seed=RS):\n",
" rng = np.random.default_rng(seed)\n",
" tr_n = tr[y[tr] == 0]\n",
" keep = np.concatenate([rng.choice(tr_n, min(N_KEEP, len(tr_n)), replace=False), tr[y[tr] != 0]])\n",
" keras.utils.set_random_seed(seed)\n",
" model = build_model()\n",
" model.compile(optimizer=keras.optimizers.Adam(learning_rate=3e-3), loss=\"sparse_categorical_crossentropy\")\n",
" t = time.time()\n",
" model.fit(X[keep], y[keep], epochs=EPOCHS, batch_size=128, verbose=0)\n",
" pred = model.predict(X[te], batch_size=2048, verbose=0).argmax(axis=1)\n",
" sens = recall_score(y[te], pred, labels=[0, 1, 2, 3], average=None, zero_division=np.nan)\n",
" return {\"acc\": np.mean(pred == y[te]),\n",
" \"macro_f1\": f1_score(y[te], pred, labels=[0, 1, 2, 3], average=\"macro\"),\n",
" **{f\"sens_{c}\": s for c, s in zip(CLASSES, sens)},\n",
" \"baseline_acc\": np.mean(y[te] == 0), # always predict N\n",
" \"n_train\": len(keep), \"train_s\": time.time() - t, \"pred\": pred, \"true\": y[te]}\n",
"\n",
"\n",
"if keras is not None:\n",
" build_model().summary()"
]
},
{
"cell_type": "markdown",
"id": "d8c831fb",
"metadata": {},
"source": [
"### 5.1 第 1 折：兩種切法各訓練一次\n",
"\n",
"同一個模型、同樣的超參數、同樣的種子，只換切法。本機 CPU 每個模型約 20 秒。"
]
},
{
"cell_type": "code",
"execution_count": 9,
"id": "93d9463b",
"metadata": {
"execution": {
"iopub.execute_input": "2026-10-06T15:12:39.069382Z",
"iopub.status.busy": "2026-10-06T15:12:39.069170Z",
"iopub.status.idle": "2026-10-06T15:13:16.311582Z",
"shell.execute_reply": "2026-10-06T15:13:16.310676Z"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"A: random beats: trained on 15,061 beats in 18.6 s\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"B: by subject: trained on 14,668 beats in 18.5 s\n",
" A: random beats B: by subject\n",
"acc 0.924 0.873\n",
"baseline_acc 0.895 0.883\n",
"macro_f1 0.712 0.406\n",
"sens_N 0.928 0.923\n",
"sens_S 0.789 0.000\n",
"sens_V 0.943 0.944\n",
"sens_F 0.757 0.000\n"
]
}
],
"source": [
"results = {name: [] for name in splits}\n",
"if keras is not None:\n",
" for name, folds in splits.items():\n",
" results[name].append(run_fold(*folds[0]))\n",
" r = results[name][0]\n",
" print(f\"{name}: trained on {r['n_train']:,} beats in {r['train_s']:.1f} s\")\n",
" show = [\"acc\", \"baseline_acc\", \"macro_f1\"] + [f\"sens_{c}\" for c in CLASSES]\n",
" print(pd.DataFrame({k: v[0] for k, v in results.items()}).loc[show].astype(float).round(3))\n",
"else:\n",
" print(\"Skipped: Keras is not available.\")"
]
},
{
"cell_type": "markdown",
"id": "ba65314c",
"metadata": {},
"source": [
"`baseline_acc` 是「全部猜 N」的多數類基準。看準確率之外，請特別看 **S 與 F 的敏感度**：同一個模型，換成依受試者切之後，這兩類的表現落差最大。\n",
"\n",
"下面的混淆矩陣把差別畫出來（列＝真實類別，欄＝預測；每列已換算成比例）："
]
},
{
"cell_type": "code",
"execution_count": 10,
"id": "96d77569",
"metadata": {
"execution": {
"iopub.execute_input": "2026-10-06T15:13:16.313984Z",
"iopub.status.busy": "2026-10-06T15:13:16.313771Z",
"iopub.status.idle": "2026-10-06T15:13:16.496467Z",
"shell.execute_reply": "2026-10-06T15:13:16.495683Z"
}
},
"outputs": [
{
"data": {
"image/png": "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",
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"if keras is not None:\n",
" fig, axes = plt.subplots(1, 2, figsize=(10, 4))\n",
" for ax, (name, res) in zip(axes, results.items()):\n",
" cm = confusion_matrix(res[0][\"true\"], res[0][\"pred\"], labels=[0, 1, 2, 3])\n",
" cm_row = cm / cm.sum(axis=1, keepdims=True)\n",
" ax.imshow(cm_row, cmap=\"Blues\", vmin=0, vmax=1)\n",
" for i in range(4):\n",
" for j in range(4):\n",
" ax.text(j, i, f\"{cm_row[i, j]:.2f}\\n({cm[i, j]})\", ha=\"center\", va=\"center\", fontsize=8,\n",
" color=\"white\" if cm_row[i, j] > 0.5 else \"black\")\n",
" ax.set_xticks(range(4), CLASSES)\n",
" ax.set_yticks(range(4), CLASSES)\n",
" ax.set_xlabel(\"predicted\")\n",
" ax.set_ylabel(\"true\")\n",
" ax.set_title(f\"{name} (fold 1)\")\n",
" plt.tight_layout()\n",
" plt.show()"
]
},
{
"cell_type": "markdown",
"id": "6674cf22",
"metadata": {},
"source": [
"### 5.2 重複另外兩折（建議開 GPU，可跳過）\n",
"\n",
"只看一折可能是運氣。這一格把兩種切法的第 2、3 折也跑完（再訓練 4 個模型），報告 3 折的平均與範圍。Colab 免費 CPU 上可能要數分鐘；跳過的話，下一格只會用第 1 折。"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "9385ec7d",
"metadata": {
"execution": {
"iopub.execute_input": "2026-10-06T15:13:16.499108Z",
"iopub.status.busy": "2026-10-06T15:13:16.498828Z",
"iopub.status.idle": "2026-10-06T15:14:33.284957Z",
"shell.execute_reply": "2026-10-06T15:14:33.283271Z"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"A: random beats fold 2: done in 18.1 s\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"A: random beats fold 3: done in 18.6 s\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"B: by subject fold 2: done in 20.1 s\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"B: by subject fold 3: done in 19.8 s\n"
]
}
],
"source": [
"if keras is not None:\n",
" for name, folds in splits.items():\n",
" for tr, te in folds[1:]:\n",
" results[name].append(run_fold(tr, te))\n",
" print(f\"{name} fold {len(results[name])}: done in {results[name][-1]['train_s']:.1f} s\")"
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "b1d04aec",
"metadata": {
"execution": {
"iopub.execute_input": "2026-10-06T15:14:33.289624Z",
"iopub.status.busy": "2026-10-06T15:14:33.289018Z",
"iopub.status.idle": "2026-10-06T15:14:33.333850Z",
"shell.execute_reply": "2026-10-06T15:14:33.332598Z"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"per fold:\n",
" split fold acc baseline_acc macro_f1 sens_N sens_S sens_V sens_F\n",
"A: random beats 1 0.924 0.895 0.712 0.928 0.789 0.943 0.757\n",
"A: random beats 2 0.954 0.895 0.782 0.963 0.746 0.957 0.652\n",
"A: random beats 3 0.967 0.895 0.796 0.979 0.657 0.969 0.623\n",
" B: by subject 1 0.873 0.883 0.406 0.923 0.000 0.944 0.000\n",
" B: by subject 2 0.938 0.914 0.450 0.970 0.007 0.838 0.148\n",
" B: by subject 3 0.742 0.886 0.364 0.738 0.282 0.971 0.000\n",
"\n",
"mean [min, max] over folds:\n",
"A: random beats: acc 0.948 [0.924, 0.967] | baseline_acc 0.895 [0.895, 0.895] | macro_f1 0.763 [0.712, 0.796] | sens_N 0.957 [0.928, 0.979] | sens_S 0.731 [0.657, 0.789] | sens_V 0.956 [0.943, 0.969] | sens_F 0.677 [0.623, 0.757]\n",
"B: by subject: acc 0.851 [0.742, 0.938] | baseline_acc 0.894 [0.883, 0.914] | macro_f1 0.407 [0.364, 0.450] | sens_N 0.877 [0.738, 0.970] | sens_S 0.096 [0.000, 0.282] | sens_V 0.918 [0.838, 0.971] | sens_F 0.049 [0.000, 0.148]\n",
"RESULTS_JSON [{\"split\": \"A: random beats\", \"fold\": 1, \"acc\": 0.924, \"baseline_acc\": 0.8948, \"macro_f1\": 0.7117, \"sens_N\": 0.9281, \"sens_S\": 0.7886, \"sens_V\": 0.9435, \"sens_F\": 0.7566}, {\"split\": \"A: random beats\", \"fold\": 2, \"acc\": 0.9544, \"baseline_acc\": 0.8948, \"macro_f1\": 0.7818, \"sens_N\": 0.9633, \"sens_S\": 0.7465, \"sens_V\": 0.9568, \"sens_F\": 0.6517}, {\"split\": \"A: random beats\", \"fold\": 3, \"acc\": 0.9668, \"baseline_acc\": 0.8948, \"macro_f1\": 0.7962, \"sens_N\": 0.9792, \"sens_S\": 0.657, \"sens_V\": 0.9692, \"sens_F\": 0.6231}, {\"split\": \"B: by subject\", \"fold\": 1, \"acc\": 0.8728, \"baseline_acc\": 0.8826, \"macro_f1\": 0.4059, \"sens_N\": 0.9225, \"sens_S\": 0.0, \"sens_V\": 0.9445, \"sens_F\": 0.0}, {\"split\": \"B: by subject\", \"fold\": 2, \"acc\": 0.9382, \"baseline_acc\": 0.9138, \"macro_f1\": 0.45, \"sens_N\": 0.9695, \"sens_S\": 0.007, \"sens_V\": 0.8385, \"sens_F\": 0.1481}, {\"split\": \"B: by subject\", \"fold\": 3, \"acc\": 0.7421, \"baseline_acc\": 0.8857, \"macro_f1\": 0.3643, \"sens_N\": 0.7379, \"sens_S\": 0.2817, \"sens_V\": 0.9714, \"sens_F\": 0.0}]\n"
]
}
],
"source": [
"if keras is not None:\n",
" import json\n",
" rows = []\n",
" for name, res in results.items():\n",
" for i, r in enumerate(res):\n",
" rows.append({\"split\": name, \"fold\": i + 1, **{k: v for k, v in r.items() if k not in (\"pred\", \"true\")}})\n",
" df = pd.DataFrame(rows)\n",
" cols = [\"acc\", \"baseline_acc\", \"macro_f1\", \"sens_N\", \"sens_S\", \"sens_V\", \"sens_F\"]\n",
" print(\"per fold:\")\n",
" print(df[[\"split\", \"fold\"] + cols].round(3).to_string(index=False))\n",
" print(\"\\nmean [min, max] over folds:\")\n",
" summary = df.groupby(\"split\")[cols].agg([\"mean\", \"min\", \"max\"])\n",
" for name in results:\n",
" print(name + \": \" + \" | \".join(\n",
" f\"{c} {summary.loc[name, (c, 'mean')]:.3f} [{summary.loc[name, (c, 'min')]:.3f}, \"\n",
" f\"{summary.loc[name, (c, 'max')]:.3f}]\" for c in cols))\n",
" # machine-readable line used by scripts/figs_ch14.py to draw the chapter figure\n",
" print(\"RESULTS_JSON\", json.dumps(df[[\"split\", \"fold\"] + cols].round(4).to_dict(orient=\"records\")))"
]
},
{
"cell_type": "markdown",
"id": "756207e6",
"metadata": {},
"source": [
"## 6. 解讀\n",
"\n",
"- **切法 A（隨機切心跳）看起來很好**：測試集裡每個人都在訓練集出現過，模型只要「認得這個人的心跳長相」就能答對，不需要學會「什麼樣的波形是上心室早期收縮」。\n",
"- **切法 B（依受試者切）比較接近部署時的情境**（但仍是同一個資料庫，不等於外部驗證）：模型要面對從沒看過的病人。S 與 F 的敏感度通常大幅下降：S 類（例如心房早期收縮）的波形本身和正常心跳很像，主要靠「來得太早」（RR 間期）辨認，而我們的切窗只有單一心跳的形狀；F 類集中在少數人，換人就幾乎認不出來。\n",
"- **準確率會騙人**：N 類占近九成，「全部猜 N」就有約 0.9 的準確率。依受試者切時，模型的準確率可能接近、甚至低於這個基準；要看 macro-F1 與各類別敏感度才看得出問題。\n",
"- 這些數字只來自 MIT-BIH 一個資料庫、44 段紀錄、單一導程，3 折之間的差距很大，請當成「方向」而不是精確估計。真正要回答「換一家醫院還準不準」，需要在另一個資料庫做外部驗證。"
]
},
{
"cell_type": "code",
"execution_count": 13,
"id": "c42da285",
"metadata": {
"execution": {
"iopub.execute_input": "2026-10-06T15:14:33.337839Z",
"iopub.status.busy": "2026-10-06T15:14:33.337426Z",
"iopub.status.idle": "2026-10-06T15:14:33.342962Z",
"shell.execute_reply": "2026-10-06T15:14:33.341646Z"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"total notebook time: 125.2 s (download + reading 1.7 s, beat preparation 1.5 s)\n"
]
}
],
"source": [
"print(f\"total notebook time: {time.time() - T_START:.1f} s \"\n",
" f\"(download + reading {T_LOAD:.1f} s, beat preparation {T_PREP:.1f} s)\")"
]
},
{
"cell_type": "markdown",
"id": "d6347ce4",
"metadata": {},
"source": [
"## 7. 動手試試\n",
"\n",
"1. **加上類別權重**：在 `model.fit` 加入 `class_weight`（例如 `{c: len(keep) / (4 * np.sum(y[keep] == c)) for c in range(4)}`），看 S、F 的敏感度與 N 的敏感度如何交換。\n",
"2. **換成循環神經網路（GRU）**：把 `build_model` 裡的卷積層換成下面這段，比較兩種切法下的差距是否仍然存在（CPU 上會慢很多，建議開 GPU）：\n",
" ```python\n",
" keras.layers.Conv1D(16, 7, padding=\"same\", activation=\"relu\"),\n",
" keras.layers.MaxPooling1D(2),\n",
" keras.layers.GRU(32), # reads the sequence step by step, keeps a hidden state\n",
" keras.layers.Dense(4, activation=\"softmax\"),\n",
" ```\n",
"3. **加長切窗**：把 `PRE, POST` 改成 `270, 270`（R 峰前後各 0.75 秒，通常會包含前後一個心跳），再把 `STEP` 改成 3。模型多看到「前一個心跳在哪裡」之後，依受試者切的 S 類敏感度有沒有改善？\n",
"4. **只分 N 與 V 兩類**：把 S、F 的心跳拿掉，重新比較兩種切法。差距變大還是變小？為什麼？\n",
"\n",
"## 參考資料\n",
"\n",
"- Moody GB, Mark RG. The impact of the MIT-BIH Arrhythmia Database. *IEEE Eng Med Biol Mag* 2001;20(3):45–50.\n",
"- Goldberger AL, et al. PhysioBank, PhysioToolkit, and PhysioNet. *Circulation* 2000;101(23):e215–e220.\n",
"- MIT-BIH Arrhythmia Database v1.0.0：（ODC-By v1.0）"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.12.2"
}
},
"nbformat": 4,
"nbformat_minor": 5
}