Error on tensorflow: Shape must be rank 2 but is rank 1 for 'MatMul_25' - tensorflow

Question

I'm trying to create a conditional GAN. However, i'm stuck as to why no matter what i do, it appears the same error over and over again.
Here's the code:
image_dim = 784 #28 * 28
Y_dimension = 10
gen_hidd_dim = 256
disc_hidd_dim = 256
z_noise_dim =100 #input noise datapoint
def xavier_init(shape):
return tf.random_normal(shape = shape, stddev = 1/tf.sqrt(shape[0]/2.0))
weights = {
'disc_H' : tf.Variable(xavier_init([image_dim + Y_dimension, disc_hidd_dim])),
'disc_final' : tf.Variable(xavier_init([disc_hidd_dim, 1])),
'gen_H': tf.Variable([z_noise_dim + Y_dimension, gen_hidd_dim]),
'gen_final': tf.Variable(xavier_init([gen_hidd_dim, image_dim]))
}
bias = {
'disc_H': tf.Variable(xavier_init([disc_hidd_dim])),
'disc_final': tf.Variable(xavier_init([1])),
'gen_H': tf.Variable(xavier_init([gen_hidd_dim])),
'gen_final': tf.Variable(xavier_init([image_dim]))
}
Z_input = tf.placeholder(tf.float32, shape= [None, z_noise_dim ], name = 'input_noise')
Y_input = tf.placeholder(tf.float32, shape= [None, Y_dimension], name='Labels')
X_input = tf.placeholder(tf.float32, shape=[None, image_dim], name = 'real_input')
def Discriminator(x,y):
inputs = tf.concat(axis = 1, values = [x,y])
hidden_layer = tf.nn.relu(tf.add(tf.matmul(inputs, weights['disc_H']), bias['disc_H']))
final_layer = tf.add(tf.matmul(hidden_layer, weights['disc_final']), bias['disc_final'])
disc_output = tf.nn.sigmoid(final_layer)
return final_layer, disc_output
def Generator(x,y):
inputs = tf.concat(axis=1, values=[x,y])
hidden_layer = tf.nn.relu(tf.add(tf.matmul(tf.cast(inputs, tf.float32), tf.cast(weights['gen_H'], tf.float32)), tf.cast(bias['gen_H'],tf.float32)))
final_layer = tf.add(tf.matmul(hidden_layer, weights['gen_final']), bias['gen_final'])
gen_output = tf.nn.sigmoid(final_layer)
return gen_output
output_Gen = Generator(Z_input, Y_input)
Right after executing the Generator i get the following error:
ValueError: Shape must be rank 2 but is rank 1 for 'MatMul_25' (op: 'MatMul') with input shapes: [?,110], [2].
What to do?

Answer 1

I think you just missed one call to xavier_init() when initialising your weights.
You have this:
weights = {
'disc_H' : tf.Variable(xavier_init([image_dim + Y_dimension, disc_hidd_dim])),
'disc_final' : tf.Variable(xavier_init([disc_hidd_dim, 1])),
'gen_H': tf.Variable([z_noise_dim + Y_dimension, gen_hidd_dim]),
'gen_final': tf.Variable(xavier_init([gen_hidd_dim, image_dim]))
}
but I think you want this:
weights = {
'disc_H' : tf.Variable(xavier_init([image_dim + Y_dimension, disc_hidd_dim])),
'disc_final' : tf.Variable(xavier_init([disc_hidd_dim, 1])),
'gen_H': tf.Variable(xavier_init([z_noise_dim + Y_dimension, gen_hidd_dim])),
'gen_final': tf.Variable(xavier_init([gen_hidd_dim, image_dim]))
}
The error message was because weights['gen_H'] had shape [2] whereas you expected it to have shape [110, 256]. This meant that the call to tf.matmul() failed because it's impossible to matrix multiply a matrix with shape [m, 110] by a matrix of shape [2]