Label scalar into one hot in Tensorr flow code?

Question

Can some one please guide me that what is meant by conversion from scalar to one hot? and what is the purpose of labels_dense.shape[0] and at the end why label one hot.flat is equal to one? def dense_to_one_hot(labels_dense, num_classes=10): """Convert class labels from scalars to one-hot vectors""" num_labels = labels_dense.shape[0] index_offset = np.arange(num_labels) * num_classes labels_one_hot = np.zeros((num_labels, num_classes)) labels_one_hot.flat[index_offset + labels_dense.ravel()] = 1 return labels_one_hot

Answer 1

I think you might find this answer describing how one hot encoding works in machine learning helpful: One Hot Encoding for Machine learning

Answer 2

I came across that same function and wrote a simpler one to understand. I am using the digits 0 to 4 which represent 5 classes.

What is the purpose of labels_dense.shape[0] ?

It returns the number of labels which is '10' in this example.

What does this code mean ?

labels_one_hot.flat[index_offset + labels_dense.ravel()] = 1

It is logic to put '1' in the correct position as you can see in the output. It is just calculating the position from the beginning of the entire one-hot representation.

So to represent the digit '0' as a one-hot vector the 45th position should be'1'. This corresponds to the 0th element of the last vector.

So

[1. 0. 0. 0. 0.]

is the one-hot representation for the digit '0' when we have 5 classes.

def onehot():

    labels_dense = numpy.array([1,2,3,4,3,4,3,2,1,0])

    print('Shape of labels_dense is ' + str(labels_dense.shape))

    index_offset = numpy.arange(10) * 5

    print('Index offset is \n' + str(index_offset))

    labels_one_hot = numpy.zeros((10, 5))

    print('index_offset + labels_dense.ravel() is\n' + str(index_offset + labels_dense.ravel()))

    labels_one_hot.flat[index_offset + labels_dense.ravel()] = 1

    print('One-hot labels are ' + str(labels_one_hot))

Output is this.

Shape of labels_dense is (10,)
Index offset is 
[ 0  5 10 15 20 25 30 35 40 45]
index_offset + labels_dense.ravel() is
[ 1  7 13 19 23 29 33 37 41 45]
One-hot labels are 

[[0. 1. 0. 0. 0.]
 [0. 0. 1. 0. 0.]
 [0. 0. 0. 1. 0.]
 [0. 0. 0. 0. 1.]
 [0. 0. 0. 1. 0.]
 [0. 0. 0. 0. 1.]
 [0. 0. 0. 1. 0.]
 [0. 0. 1. 0. 0.]
 [0. 1. 0. 0. 0.]
 [1. 0. 0. 0. 0.]]