【python實現卷積神經網絡】Flatten層實現

代碼來源：https://github.com/eriklindernoren/ML-From-Scratch

卷積神經網絡中卷積層Conv2D（帶stride、padding）的具體實現：https://www.cnblogs.com/xiximayou/p/12706576.html

激活函數的實現（sigmoid、softmax、tanh、relu、leakyrelu、elu、selu、softplus）：https://www.cnblogs.com/xiximayou/p/12713081.html

損失函數定義（均方誤差、交叉熵損失）：https://www.cnblogs.com/xiximayou/p/12713198.html

優化器的實現（SGD、Nesterov、Adagrad、Adadelta、RMSprop、Adam）：https://www.cnblogs.com/xiximayou/p/12713594.html

卷積層反向傳播過程：https://www.cnblogs.com/xiximayou/p/12713930.html

全連接層實現：https://www.cnblogs.com/xiximayou/p/12720017.html

批量歸一化層實現：https://www.cnblogs.com/xiximayou/p/12720211.html

池化層實現：https://www.cnblogs.com/xiximayou/p/12720324.html

padding2D實現：https://www.cnblogs.com/xiximayou/p/12720454.html

 

這就相當於是pytorch中的在全連接層之前使用view()函數類似的操作：

class Flatten(Layer):
    """ Turns a multidimensional matrix into two-dimensional """
    def __init__(self, input_shape=None):
        self.prev_shape = None
        self.trainable = True
        self.input_shape = input_shape

    def forward_pass(self, X, training=True):
        self.prev_shape = X.shape
        return X.reshape((X.shape[0], -1))

    def backward_pass(self, accum_grad):
        return accum_grad.reshape(self.prev_shape)

    def output_shape(self):
        return (np.prod(self.input_shape),)

需要注意反向傳播時的形狀的改變。

還有Reshape層：

class Reshape(Layer):
    """ Reshapes the input tensor into specified shape
    Parameters:
    -----------
    shape: tuple
        The shape which the input shall be reshaped to.
    """
    def __init__(self, shape, input_shape=None):
        self.prev_shape = None
        self.trainable = True
        self.shape = shape
        self.input_shape = input_shape

    def forward_pass(self, X, training=True):
        self.prev_shape = X.shape
        return X.reshape((X.shape[0], ) + self.shape)

    def backward_pass(self, accum_grad):
        return accum_grad.reshape(self.prev_shape)

    def output_shape(self):
        return self.shape