From fd232d65fa42219e2bee0491396bd52004975b4b Mon Sep 17 00:00:00 2001
From: kingecg <kingecg@qq.com>
Date: Tue, 30 Dec 2025 23:23:51 +0800
Subject: [PATCH] =?UTF-8?q?```=20feat(examples):=20=E6=9B=B4=E6=96=B0CNN?=
 =?UTF-8?q?=E7=A4=BA=E4=BE=8B=E4=B8=AD=E7=9A=84=E5=85=A8=E8=BF=9E=E6=8E=A5?=
 =?UTF-8?q?=E5=B1=82=E5=AE=9E=E7=8E=B0?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 修复权重矩阵大小计算，根据展平后的实际大小动态创建权重
- 添加张量重塑逻辑，将展平后的张量转换为正确的2D格式进行矩阵乘法
- 使用动态生成的权重值替代固定的权重数组
- 确保矩阵乘法维度匹配：(1, N) * (N, 2) = (1, 2)
```
---
 examples/cnn_example.go | 23 +++++++++++++++++++----
 1 file changed, 19 insertions(+), 4 deletions(-)

diff --git a/examples/cnn_example.go b/examples/cnn_example.go
index f2f08e6..30a06f7 100644
--- a/examples/cnn_example.go
+++ b/examples/cnn_example.go
@@ -65,15 +65,30 @@ func main() {
 	fmt.Printf("展平后大小: %d\n", flattened.Size())
 	
 	// 创建一些随机权重进行全连接层操作
-	weightsData := []float64{0.1, 0.2, 0.3, 0.4, 0.5, 0.6}
-	weightsShape := []int{flattened.Size(), 2} // 输出2类（猫/狗）
+	// 由于池化后是2x2，展平后应该是4个元素，所以我们需要4x2的权重矩阵
+	flattenedSize := flattened.Size()
+	weightsData := make([]float64, flattenedSize * 2) // flattenedSize*2个权重值
+	for i := range weightsData {
+		weightsData[i] = 0.1 * float64(i+1) // 填充一些递增的值
+	}
+	weightsShape := []int{flattenedSize, 2} // 输出2类（猫/狗）
 	weights, err := gotensor.NewTensor(weightsData, weightsShape)
 	if err != nil {
 		panic(err)
 	}
 	
-	// 全连接层计算 (矩阵乘法)
-	fcResult, err := flattened.MatMul(weights)
+	// 重塑flattened张量为2D格式以进行矩阵乘法
+	reshapedFlattenedData := make([]float64, flattenedSize)
+	for i := 0; i < flattenedSize; i++ {
+		reshapedFlattenedData[i], _ = flattened.Data.Get(i)
+	}
+	reshapedFlattened, err := gotensor.NewTensor(reshapedFlattenedData, []int{1, flattenedSize}) // 作为1xN的矩阵
+	if err != nil {
+		panic(err)
+	}
+	
+	// 全连接层计算 (矩阵乘法) - 现在是 (1, N) * (N, 2) = (1, 2)
+	fcResult, err := reshapedFlattened.MatMul(weights)
 	if err != nil {
 		panic(err)
 	}