Suppose I have the input of form [a,b,c] and I want to produce the output [a * b, b * c]
I thus would try something like this:
input = Input((3,))
output = Lambda(lambda x: [x[0]*x[1], x[1]*x[2]], output_shape = (2,))(input)
model = Model(input, output)
However, it seems to not work. When I print the summary I get:
shape(input) = (None,3)
shape(output) = [(3,),(3,)] ## shouldn't this be (None,2)?
Input tensor x in your case has shape (None,3) so x[0] is first sample in batch and not a first feature. You should use x[:,0] instead. Also returing list of tensors from layers means that layers has multiple outputs (not single output with multiple features) so you will need to stack them with tf.stack.
Here is sample code:
input = Input((3,))
output = Lambda(lambda x: tf.stack([x[:,0]*x[:,1], x[:,1]*x[:,2]], axis=1))(input)
Results
input.shape # TensorShape([None, 3])
output.shape # TensorShape([None, 2])
output = Lambda(lambda x: x[:,0:2]*x[:,1:3])(input)