TensorFlow InvalidArgumentError: Matrix size-compatible: In[0]: [100,784], In[1]: [500,10]

Question

I'm new to tensorflow and am following a tutorial. I am getting an error that says:

InvalidArgumentError (see above for traceback): Matrix size-compatible: In[0]: [100,784], In[1]: [500,10]
     [[Node: MatMul_3 = MatMul[T=DT_FLOAT, transpose_a=false, transpose_b=false, _device="/job:localhost/replica:0/task:0/cpu:0"](_recv_Placeholder_0, Variable_6/read)]]

Here is my code:

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

mnist = input_data.read_data_sets("/tmp/data/", one_hot=True)

n_nodes_hl1 = 500
n_nodes_hl2 = 500
n_nodes_hl3 = 500

n_classes = 10
batch_size = 100

x = tf.placeholder('float') #this second parameter makes sure that the image fed in is 28*28
y = tf.placeholder('float')

def neural_network_model(data):
    hidden_1_layer = {'weights':tf.Variable(tf.random_normal([784, n_nodes_hl1])), 'biases':tf.Variable(tf.random_normal([n_nodes_hl1]))}
    hidden_2_layer = {'weights':tf.Variable(tf.random_normal([n_nodes_hl1, n_nodes_hl2])), 'biases':tf.Variable(tf.random_normal([n_nodes_hl2]))}
    hidden_3_layer = {'weights':tf.Variable(tf.random_normal([n_nodes_hl2, n_nodes_hl3])), 'biases':tf.Variable(tf.random_normal([n_nodes_hl3]))}
    output_layer = {'weights':tf.Variable(tf.random_normal([n_nodes_hl3, n_classes])), 'biases':tf.Variable(tf.random_normal([n_classes]))}

    # input_data * weights + biases
    l1 = tf.add(tf.matmul(data, hidden_1_layer['weights']), hidden_1_layer['biases'])
    # activation function
    l1 = tf.nn.relu(l1)

    l2 = tf.add(tf.matmul(data, hidden_2_layer['weights']), hidden_2_layer['biases'])
    l2 = tf.nn.relu(l2)

    l3 = tf.add(tf.matmul(data, hidden_3_layer['weights']), hidden_3_layer['biases'])
    l3 = tf.nn.relu(l3)

    output = tf.matmul(data, output_layer['weights']) + output_layer['biases']
    return output

def train_neural_network(x):
    prediction = neural_network_model(x)
    cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(prediction, y))
    #learning rate = 0.001
    optimizer = tf.train.AdamOptimizer().minimize(cost)
    hm_epochs = 10
    with tf.Session() as sess:
        sess.run(tf.initialize_all_variables())
        for epoch in range(hm_epochs):
            epoch_loss = 0
            for _ in range(int(mnist.train.num_examples/batch_size)):
                epoch_x, epoch_y = mnist.train.next_batch(batch_size)
                _, c = sess.run([optimizer, cost], feed_dict={x:epoch_x,y:epoch_y})//THIS IS THE LINE WHERE THE ERROR 0CCURS
                epoch_loss += c
            print 'Epoch ' + epoch + ' completed out of ' + hm_epoch + ' loss: ' + epoch_loss
        correct = tf.equal(tf.argmax(prediction,1), tf.argmax(y,1))
        accuracy = tf.reduce_mean(tf.cast(correct, 'float'))
        print 'Accuracy: ' + accuracy.eval({x:mnist.test.images, y:mnist.test.labels})

train_neural_network(x)

I have marked the line where the error occurs what am I doing wrong and how can I fix it?

Stack overflow wants me to write more, it says that there is not enough details and too much code. I honestly don't understand tensorflow well enough to add any more details. I'm hoping someone can just help me with this. I think the problem is that optimizer and cost have different dimensions, but I don't understand why or what I should do about it.

Answer 1

One error lies in this line

 l2 = tf.add(tf.matmul(data, hidden_2_layer['weights']), hidden_2_layer['biases'])

Your second weights variable has dimensions 500 x 500, but your data variable was fed with data 100x784 so multiplication is incompatible. Make this,

 l2 = tf.add(tf.matmul(l1, hidden_2_layer['weights']), hidden_2_layer['biases'])

Also make the corresponding change for l3 and output.

---

Always specify a shape for the placeholder, like this,

x = tf.placeholder(tf.float32, shape=(None, 784))

This will allow you to catch such errors while building the graph and TensorFlow will be able to pinpoint these errors.