我必须用keras训练一个神经网络 . 为此，我使用了一些具有以下形状的测试数据：
print(" Training data: {}".format(x_Train.shape))
print(" Training data: {}".format(y_Train.shape))
print(" Test data: {}".format(x_Test.shape))
print(" Test data: {}".format(y_Test.shape))
....
Training data: (128, 90, 561)
Training data: (128,)
Test data: (43, 90, 561)
Test data: (43,)
而这个网络架构：
class NeuralNetwork:
@staticmethod
def Build(Width, Depth, Classes, Drop = 0.5):
Model = Sequential()
Model.add(Conv1D(filters = 32,
kernel_size = 5,
input_shape = (Width, Depth)
))
Model.add(Activation("relu"))
Model.add(MaxPooling1D(pool_size = 2,
strides = 2
))
Model.add(Conv1D(filters = 64,
kernel_size = 3
))
Model.add(Activation("relu"))
Model.add(MaxPooling1D(pool_size = 2,
strides = 2
))
Model.add(Flatten())
Model.add(Dense(1024))
Model.add(Dropout(Drop))
Model.add(Dense(Classes))
Model.add(Activation("softmax"))
return Model
但是当我尝试训练我的模型时，我遇到了这个错误：
ValueError: Error when checking target: expected activation_3 to have shape (12,) but got array with shape (1,)
我使用此代码进行培训：
print("[INFO] Train model...")
self.__Model = NeuralNetwork.Build(90, 561, 12)
plot_model(self.__Model, show_layer_names = True, show_shapes = True)
self.__Model.compile(loss = "binary_crossentropy", optimizer = Adam(lr = self.__Learnrate), metrics = ["accuracy"])
self.__Model.fit(x_Train,
y_Train,
validation_data = (x_Test, y_Test),
batch_size = self.__BatchSize,
epochs = self.__Epochs,
verbose = 1
)
我没有得到这个错误的来源 . 我用tensorflow测试整个代码，它工作正常 . 但我对keras的重新设计做错了 .
谢谢你的提示或其他东西......
1 回答
看起来你混淆了目标和你的损失功能 . 我猜你的目标是你 class 的整数标签
y = [3, 5, 6, ...]你最多有12个 class . 在这种情况下，您的损失应该是sparse_categorical_crossentropy，因为您想要预测12个互斥类中的1个 .该错误表示您输出的分布超过12个类但只提供单个目标 . 像
out = [0.2, 0.5, 0.1, ...]和y = [2]这样的东西是（12，）和（1，）之间的形状不匹配 . 稀疏分类将您的目标标签转换为单热矢量，因此它变为y = [0,0,1,0,...]