CODEWITHSUNDEEP
pythonmathtensorflowlinear-algebraalgebra
garciparedes
garciparedes

Reputation: 1847

Why TensorFlow's division hasn't have axis parameter?

My problem is as follows:

When I want to divide a matrix by a vector in column axis, but TensorFlow only provides row division when A is a matrix with [n,n] shape and v a vector with [1,n] shape.

My solution was this:

tf.transpose(tf.div(tf.transpose(A), v))

I tried this but it doesn't works: Update: It Works! 🙂

tf.div(A, tf.transpose(v))

So my question is if exists more efficient solution for this problem.

Upvotes: 1

Views: 1970

Answers (1)

Salvador Dali
Salvador Dali

Reputation: 222571

Your second solution with A / tf.transpose(v) should work. tf.div() does not have an axis parameter because it "Divides x / y elementwise". So both tensors should have the same dimensions.

In your case it works with different dimensions because tf.div supports broadcasting. More about broadcasting is here.

Also it is as efficient as it could be, so no need to look for anything else.

Upvotes: 1

Related Questions