Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

commented code in loss functions #9

Open
khums opened this issue Jul 3, 2019 · 0 comments
Open

commented code in loss functions #9

khums opened this issue Jul 3, 2019 · 0 comments

Comments

@khums
Copy link

khums commented Jul 3, 2019

I have a few queries regarding the loss function implementation:
a) the lambda value in loss function is set to 0 therefore the l2_norm is not considered in the loss equation
b)The embedding loss is just the distance between the two embedding spaces as per the code:
return tf.reduce_mean(tf.square(Fx - Fe)) #tf.trace(tf.matmul(C1, tf.transpose(C1))) + self.config.solver.lagrange_const * tf.trace(tf.matmul(C2, tf.transpose(C2))) + self.config.solver.lagrange_const * tf.trace(tf.matmul(C3, tf.transpose(C3)))
this may not converge to the optimal point
c) The output loss is (1/y1 * y0) whereas the equation in the paper says (1/(y1 * y0)) resulting in a different alpha value for the output loss function.

the output loss is similar to bp-mll loss and there are different implementations already available for the loss function. I came across a paper that cited this implementation and the implementation here relates more to a DistAE rather than CorrAE as described in one of Wang et. al.'s paper https://ttic.uchicago.edu/~wwang5/papers/icml15a.pdf

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant