Cannot feed value of shape (32, 1, 3) for Tensor u'/X:0', which has shape '(?, 3)

Question

I want to create a reinforcement project, but I struggle with some problems.

I have a class for my neural network. Consisting of one Input Layer, two Hidden Layer and one output-Layer. It is created with tflearn.

class Network():
    self.inputs, self.outputs = self.createNetwork()
    [...]
    def createNetwork(self):

        # Input-Layer
        inputs = tflearn.input_data(shape=[None, 3])

        # Hidden-Layer L1
        net = tflearn.fully_connected(inputs, 400, activation='relu')                                   

        # Hidden Layer L2
        net = tflearn.fully_connected(net, 300, activation='relu')                                   

        # Final layer weights are init to Uniform[-3e-3, 3e-3]
        weight_init_final = tflearn.initializations.uniform(minval=-0.003, maxval=0.003)
        output = tflearn.fully_connected(net, 1, activation='tanh', weights_init=weight_init_final)

        return inputs, output

And the method for predicting a value

def predict(self, inputs):
   return self.sess.run(self.outputs, feed_dict={
   self.inputs: inputs
   })

I train with a batch size of 32 and want to predict a value.

network.predict(test_batch)

Sadly I get an error 'Cannot feed value of shape (32, 1, 3) for Tensor u'/X:0', which has shape '(?, 3)' test_batch.shape results in (32, 1, 3)

Answer 1

Your input layer is expecting a shape of (?,3) --> inputs = tflearn.input_data(shape=[None, 3])
The input_shape should be (None, 1, 3) to match the shape of your data.