Method forward

code/chapter9.py:633–676

池化层前向传播 参数说明： X：输入数组，形状为 (n_samp, in_rows, in_cols, in_ch) retain_derived：是否保留中间变量，以便反向传播时再次使用，bool型 输出说明： Y：输出结果，形状为 (n_samp, out_rows, out_cols, out_ch)

(self, X, retain_derived=True)

        self.is_initialized = True

    def forward(self, X, retain_derived=True):
        """
        池化层前向传播

        参数说明：
        X：输入数组，形状为 (n_samp, in_rows, in_cols, in_ch)
        retain_derived：是否保留中间变量，以便反向传播时再次使用，bool型
        
        输出说明：
        Y：输出结果，形状为 (n_samp, out_rows, out_cols, out_ch)
        """
        if not self.is_initialized:
            self.in_ch = self.out_ch = X.shape[3]
            self._init_params()

        n_samp, in_rows, in_cols, nc_in = X.shape
        (fr, fc), s, p = self.kernel_shape, self.stride, self.pad
        X_pad, (pr1, pr2, pc1, pc2) = pad2D(X, p, self.kernel_shape, s)

        out_rows = int((in_rows + pr1 + pr2 - fr) / s + 1)
        out_cols = int((in_cols + pc1 + pc2 - fc) / s + 1)

        if self.mode == "max":
            pool_fn = np.max
        elif self.mode == "average":
            pool_fn = np.mean

        Y = np.zeros((n_samp, out_rows, out_cols, self.out_ch))
        for m in range(n_samp):
            for i in range(out_rows):
                for j in range(out_cols):
                    for c in range(self.out_ch):
                        i0, i1 = i * s, (i * s) + fr
                        j0, j1 = j * s, (j * s) + fc

                        xi = X_pad[m, i0:i1, j0:j1, c]
                        Y[m, i, j, c] = pool_fn(xi)

        if retain_derived:
            self.X.append(X)
            self.derived_variables["out_rows"].append(out_rows)
            self.derived_variables["out_cols"].append(out_cols)

        return Y

    def backward(self, dLdy, retain_grads=True):
        """