CODEWITHSUNDEEP
pythonnumpytensorflow
G4bri3l
G4bri3l

Reputation: 5146

How do I multiply a matrix by a scalar only for given indices in Tensorflow?

I am trying to find an efficient way to multiply specific values within a matrix for a given scalar. Let's see a quick example.

Given a matrix M of values between 1 and 10 like so:

enter image description here

I want to multiply every cell that has value smaller than 3, by 2. Now I know I can find the coordinates of all items that have value 1 in tensorflow with tf.where(M < 3) but I am struggling to find a good scalable solution to attain what I want. The transformation should be something like this:

enter image description here

How can I leverage this info to multiply only the cells at the given coordinates by 2 ?

Keep in mind that my matrices might be mich bigger than 3x3

Upvotes: 0

Views: 487

Answers (2)

G4bri3l
G4bri3l

Reputation: 5146

I found out how to do this in tensorflow without having to do any transformation from numpy to tensorflow and vice-versa.

my_matrix = tf.constant([[1, 5, 8], [2, 2, 2], [9, 7, 6]])
result = tf.where(
    tf.less(my_matrix, tf.constant(3)),
    tf.scalar_mul(2, my_matrix),
    my_matrix
)

@Josh answer helped me look into the right direction

Upvotes: 0

Josh Herzberg
Josh Herzberg

Reputation: 318

M = np.array([[1, 5, 8],[2, 2, 2], [9, 7, 6]])
M[M==1] = 2
print(M)

array([[2, 5, 8],
       [2, 2, 2],
       [9, 7, 6]])

Upvotes: 2

Related Questions