Cost function for Convolution neural network

Question

I am doing Text Classification by Convolution Neural Network. In the example MNIST they have 60.000 images examples of hand-written digits, each image has size 28 x 28 and there are 10 labels (from 0 to 9). So the size of Weight will be 784 * 10 (28 * 28 = 784)

Here is their code:

x = tf.placeholder("float", [None, 784])
W = tf.Variable(tf.zeros([784,10]))
b = tf.Variable(tf.zeros([10]))

In my case, I applied word2vec to encode my documents. The result "dictionary size" of word embedding is 2000 and the embedding size is 128. There are 45 labels. I tried to do the same as the example but It did not work. Here what I did: I treated each document same as image. For instance, The document could be represent as the matrix of 2000 x 128 (for words appear in document I appended the word Vector value for that column and left other equal zero. I have a trouble with creating W and x since my input data is a numpy array of 2000 x 128 while x = tf.placeholder("float", [None, 256000]) . The size did not match.

Could nay one suggest any advises ?

Thanks

Answer 1

Placeholder x is an array of flattened images, where first dimension None corresponds to batch size, ie number of images, and 256000 = 2000 * 128 . So, in order to feed x properly you need to flatten your input. Since you mention that your input is numpy arrays, take a look at numpy.reshape and flatten .