ValueError：检查目标时出错：预期的预测具有4维，但数组的形状为（256，1，1）

Question

我正在尝试训练模型Keras，但遇到问题：

g = ImageDataGenerator(featurewise_center=True,
                    featurewise_std_normalization=True,
                    rotation_range=45,
                    width_shift_range=0.2,
                    height_shift_range=0.2,
                    horizontal_flip=True,
                    validation_split=validation_split,
                    preprocessing_function=lambda x: x / 127 - 1)

g_train = g.flow(x_train, y_train,
             batch_size=batch_size,
             subset='training')

g_valid = g.flow(x_train, y_train,
             batch_size=batch_size,
             shuffle=False,
             subset='validation')


history = network.fit_generator(g_train,
                          steps_per_epoch=len(x_train) / 32, 
                          epochs=epochs)

ValueError: Error when checking target: expected predictions to have 4 dimensions, but got array with shape (256, 1)

有人知道为什么吗？在我看来，这很像文档中的示例。

x_train.shape
(50000, 32, 32, 1)
y_train.shape
(50000, 1, 1)