Problem Implementing "An Introduction to Implementing Neural Network Using tensorFlow"

@graydon1234 wrote:

I am following the guide on An Introduction to Implementing Neural Networks using TensorFlow by Faizan Shaikh and am having difficulties with some parts. I know I am jumping into the problem, but it is a long tutorial. I am willing to provide any other background information if needed. I am new to TensorFlow and am really trying to read all the documentation, but it is still very challenging. Please help me work through this issue.

Problem:
I receive the following error: InvalidArgumentError (see above for traceback): logits and labels must be broadcastable: logits_size=[512,10] labels_size=[128,10]

From what I have read on the discussion of this tutorial, is that this is caused by batch_x and batch_y not being the same length. I confirm that they are not the same length. However, I do not know how to get them the same length.
He created a batch_creator function which should handle this. Here is the batch_creator function:

def batch_creator(batch_size, dataset_length, dataset_name):
"""Create batch with random samples and return appropriate format"""
batch_mask = rng.choice(dataset_length, batch_size)

batch_x = eval(dataset_name + '_x')[[batch_mask]].reshape(-1, 784)
batch_x = preprocess(batch_x)

if dataset_name == 'train':
    batch_y = eval(dataset_name).ix[batch_mask, 'label'].values
    batch_y = dense_to_one_hot(batch_y)
    
return batch_x, batch_y

I do not understand the line:
batch_x = eval(dataset_name + '_x')[[batch_mask]].reshape(-1, 784)

At this point the batch_x size is incorrect [512, 784] and the batch_y size is [128,10].

I am trying my best to understand what the batch_x line is doing, but cannot seem to figure it out and do not know how to properly modify the size. I also get deprecation warnings for this line of code.

Inside the nested for loop is where batch_creator is called.

with tf.Session() as sess:
sess.run(init)

# For each epoch, do:
    # For each batch (total data rows / batch size), do:
        # Create pre-processed batch
        # Run optimizer by feeding batch
        # Find cost and reiterate to minimize 

for epoch in range(epochs):
    avg_cost = 0
    total_batch = int(train.shape[0] / batch_size)

    for batch in range(total_batch):
        batch_x, batch_y = batch_creator(batch_size, train_x.shape[0], 'train')

        _, cost = sess.run([optimizer, cost], feed_dict = {x: batch_x, y: batch_y})

If anyone could assist me with this problem, or at least the one line of code I don’t understand that would be super helpful. If my problem doesn’t make sense, someone else has posted something similar in the comments which you can look at.

Thank you

Posts: 1

Participants: 1

Read full topic