Input_shape with Conv2D error - keras

Question

I want to use the Keras Conv2D but got errors:
model.add(Conv2D(64, (2, 2), padding='valid', data_format='channels_last', input_shape=(1, 4, 4, 1)))
The Keras doc tells us that input shape is a 4D tensor, but it throws this error:
ValueError: Input 0 is incompatible with layer conv2d_1: expected ndim=4, found ndim=5.
I did some debugging and found there's a check to parameters in topology.py:
if spec.ndim is not None:
if K.ndim(x) != spec.ndim:
raise ValueError('Input ' + str(input_index) +
' is incompatible with layer ' +
self.name + ': expected ndim=' +
str(spec.ndim) + ', found ndim=' +
str(K.ndim(x)))
I found that x = Tensor("conv2d_1_input:0", shape=(?, 1, 4, 4, 1), dtype=float32) is a tensor with dim=5 and spec is an instance of InputSpec with dim=4, it never is equal. How to solve this problem?
The code :
def _build_model(self):
# Neural Net for Deep-Q learning Model
model = Sequential()
model.add(Conv2D(64, (2, 2), padding='valid', data_format='channels_last', input_shape=(1, 4, 4, 1)))
model.add(Conv2D(128, 3, strides=(1, 1), padding='valid'))
model.add(Flatten())
model.add(Dense(16, activation='relu'))
model.add(Dense(self.action_size, activation='linear'))
model.compile(loss='mse', optimizer=Adam(lr=self.learning_rate))
return model

Answer 1

Try this:
model.add(Conv2D(64, (2, 2), padding='valid', data_format='channels_last', input_shape=(4, 4, 1)))
The Convolutional2D layer expects #samples * height * width * channels. The number of samples is inferred from your model.fit() function where you feed in you data.
If you look at MNIST as the simplest example, this works:
model = Sequential()
model.add(Conv2D(32, (3, 3), padding='same', input_shape=(28, 28, 1)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
...
history = model.fit(X-train, y_train, batch_size=32, epochs=1)