Feed_dict And Numpy Reshape

say I have a tensorflow shape: y_ = tf.placeholder(tf.float32,[None,19],name='Labels') My thinking here is to get each time 19 a vector of 19 elements and add it(pluging) it to

Solution 1:

This works:

import numpy as np
import tensorflow as tf
y_ = tf.placeholder(tf.float32,[None,19],name='Labels')
sess = tf.InteractiveSession()
labels = np.zeros(57, dtype=np.float32)
sess.run(y_, feed_dict = {y_: np.reshape(labels, (3,19))})

Could it be that your inputlabel is of the wrong type?

Solution 2:

You just need to convert the list of (nSamples) integers coding the 1-hot label to an array of shape (nSamples, 19) containing only 0s and 1s when you import your data,

eg (1, 8, 2) -> [[1, 0, 0, ...], [0, 0, 1, 0, 0, ...], [0, 1, 0, 0, ...]]

You could do it like that:

label_1_hot_coding = row[-1]
array_of_bits = np.zeros(numberOFClasses)
n = 1
for i,_in enumerate(array_of_bits):
    if n== label_1_hot_coding:
        array_of_bits[i] = 1
Training_Labels.append(array_of_bits)

Your labels are now naturally of shape (nSamples, numberOfClasses), and the batches of shape (batchSize, numberOfClasses), which is what the rest of your program wants.