Weighted Sampling Preparation

Author: Sayed-Hossein-Hosseini

RhythmAttention_Hybrid_CNN_Transformer_Architecture_for_Arrhythmia_Classification.ipynb

@@ -151,6 +151,51 @@
         "print(f\"X_train shape: {X_train.shape}\") # Expected: (87554, 187, 1)\n",
         "print(f\"X_test shape: {X_test.shape}\")   # Expected: (21892, 187, 1)"
       ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "## **Weighted Sampling Preparation**"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "execution_count": 4,
+      "metadata": {},
+      "outputs": [
+        {
+          "name": "stdout",
+          "output_type": "stream",
+          "text": [
+            "Class Weights calculated for balanced learning:\n",
+            "  Class 0: 0.2416\n",
+            "  Class 1: 7.8771\n",
+            "  Class 2: 3.0254\n",
+            "  Class 3: 27.3179\n",
+            "  Class 4: 2.7229\n"
+          ]
+        }
+      ],
+      "source": [
+        "# 4. Calculate class weights to handle extreme imbalance\n",
+        "# 'balanced' mode automatically assigns higher weights to minority classes\n",
+        "weights = class_weight.compute_class_weight(\n",
+        "    class_weight='balanced',\n",
+        "    classes=np.unique(y_train),\n",
+        "    y=y_train\n",
+        ")\n",
+        "class_weights_dict = dict(enumerate(weights))\n",
+        "\n",
+        "# 5. Map weights to every individual sample for the sampler\n",
+        "sample_weights = np.array([class_weights_dict[cls] for cls in y_train])\n",
+        "# Normalize weights into probabilities\n",
+        "sample_probabilities = sample_weights / np.sum(sample_weights)\n",
+        "\n",
+        "print(\"Class Weights calculated for balanced learning:\")\n",
+        "for cls, w in class_weights_dict.items():\n",
+        "    print(f\"  Class {cls}: {w:.4f}\")"
+      ]
     }
   ],
   "metadata": {