Keras error: expected dense_input_1 to have 3 dimensions

Question

I am trying out a simple model in Keras, which I want to take as input a matrix of size 5x3. In the below example, this is specified by using input_shape=(5, 3) when adding the first dense layer.

from keras.models import Sequential
from keras.layers import Dense, Activation
from keras.optimizers import Adam
import numpy as np


model = Sequential()
model.add(Dense(32, input_shape=(5, 3)))
model.add(Activation('relu'))
model.add(Dense(32))
model.add(Activation('relu'))
model.add(Dense(4))


adam = Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)
model.compile(loss='mean_squared_error', optimizer=adam)


x = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9], [10, 11, 12], [13, 14, 15]])


y = model.predict(x)

However, when I run the code, the model.predict() function gives the following error:

ValueError: Error when checking : expected dense_input_1 to have 3 dimensions, but got array with shape (5, 3)

But I don't understand the error. The shape of x is (5, 3), and this is exactly what I have told the first dense layer to expect as input. Why is it therefore expecting three dimensions? It seems that this may be something to do with the batch size, but I thought that input_shape is only referring to the shape of the network, and is nothing to do with the batch size...

Answer 1

The problem lies here:

model.add(Dense(32, input_shape=(5, 3)))

it should be:

model.add(Dense(32, input_shape=(3,)))

This first example dimension is not included in input_shape. Also because it's actually dependent on batch_size set during network fitting. If you want to specify you could try:

model.add(Dense(32, batch_input_shape=(5, 3)))

EDIT:

From your comment I understood that you want your input to have shape=(5,3) in this case you need to:

1. reshape your x by setting:

   x = x.reshape((1, 5, 3))
   

   where first dimension comes from examples.

2. You need to flatten your model at some stage. It's because without it you'll be passing a 2d input through your network. I advice you to do the following:

   model = Sequential()
   model.add(Dense(32, input_shape=(5, 3)))
   model.add(Activation('relu'))
   model.add(Dense(32))
   model.add(Activation('relu'))
   model.add(Flatten())
   model.add(Dense(4))