CODEWITHSUNDEEP
pythontensorflowtensorflow2.0loss-function
Rishith Ellath Meethal
Rishith Ellath Meethal

Reputation: 70

Custom loss in Tensorflow 2.0.0

I want to implement a custom loss which can be calculated by using each sample. Calculation of the loss is a little complicated and requires me to use an external python file for this (or one can assume that we give the inputs to a function).

How can I implement this? Is it possible to use @tf.function annotation and make it a graph?

This is how it is supposed to look 

def loss(input,output):
    loss = 0    
    for x, y in zip(input, output):
        sim = Class(x) 
        a = sim.GetA()
        b = sim.GetB()

        loss = loss + np.linalg.norm(np.dot(a,b)+y)
    return loss

Upvotes: 1

Views: 190

Answers (1)

Rishith Ellath Meethal
Rishith Ellath Meethal

Reputation: 70

An implementation of the same via PyTorch was possible as it supports dynamic computational graph

Upvotes: 1

Related Questions