update 293

机器学习/293.实现万能近似函数: 神经网络的架构设计.md

@@ -161,32 +161,25 @@ type TrainingItem = [number[], number[]]
 
 ```ts
 class NeuralNetwork {
+  /** 获取上一层神经网络各节点的值 */
+  private getPreviousLayerValues(layerIndex: number, trainingItem: TraningItem) {
+    if (layerIndex >= 0) {
+      return this.networkStructor[layerIndex].neurals.map((neural) => neural.value);
+    }
+    return trainingItem[0];
+  }
+
   private modelFunction(trainingItem: TraningItem) {
     this.networkStructor.forEach((layer, layerIndex) => {
-      // 遍历上一层所有输入
-      const visitPreviousLayerInputs = (
-        callback: (value: number, index: number) => void
-      ) => {
-        if (layerIndex === 0) {
-          // 第一层的输入来自 Training data
-          trainingItem[0].forEach(callback);
-        } else {
-          // 之后每层输入来自于上一层
-          this.networkStructor[layerIndex - 1].neurals.forEach(
-            (neural, index) => {
-              callback(neural.value!, index);
-            }
-          );
-        }
-      };
-
       layer.neurals.forEach((neural) => {
         // 前置节点的值 * w 的总和
-        let previousWeightCount = 0;
-        visitPreviousLayerInputs((value, index) => {
-          previousWeightCount += value * neural.w[index];
-        });
-        const activateResult = activate(layer.activation)(previousWeightCount + neural.b);
+        let weightCount = 0;
+        this.getPreviousLayerValues(layerIndex - 1, trainingItem).forEach(
+          (value, index) => {
+            weightCount += value * neural.w[index];
+          },
+        );
+        const activateResult = activate(layer.activation)(weightCount + neural.b);
         neural.value = neural.c * activateResult;
       });
     });