About the implementation of the off policy corrections

[zhkmxx9302013]

Thanks for providing the code of HIRO.
I‘ve got a question about the implementation of the off-policy corrections function.
What does this comment # TODO: Doesn't include subgoal transitions!! mean, and why this function return the subgoal directly, is there anything wrong with the sampling of the candidate goal? Thanks for asking :).

[smalltingting]

Thanks for providing the code of HIRO.
I‘ve got a question about the implementation of the off-policy corrections function.
What does this comment # TODO: Doesn't include subgoal transitions!! mean, and why this function return the subgoal directly, is there anything wrong with the sampling of the candidate goal? Thanks for asking :).

I have the same problem with you. Have you solved it?

[bhairavmehta95]

Oh man - I never even saw this :( I apologize.

What does this comment # TODO: Doesn't include subgoal transitions!! mean,

So when I open sourced this code, I was actually working in a slightly different domain (images, not states) so Equation 2 in this code was never implemented (since Eq2 of subgoal transitions doesn't make sense in a latent space) So, to use this correctly, you'd have to actually transition the subgoal in the main loop.

and why this function return the subgoal directly, is there anything wrong with the sampling of the candidate goal?

Nope; Because I didn't add the above, I just added the subgoals directly. But, if we add the above, it works fine.

[zhkmxx9302013]

Oh man - I never even saw this :( I apologize.

What does this comment # TODO: Doesn't include subgoal transitions!! mean,

So when I open sourced this code, I was actually working in a slightly different domain (images, not states) so Equation 2 in this code was never implemented (since Eq2 of subgoal transitions doesn't make sense in a latent space) So, to use this correctly, you'd have to actually transition the subgoal in the main loop.

and why this function return the subgoal directly, is there anything wrong with the sampling of the candidate goal?

Nope; Because I didn't add the above, I just added the subgoals directly. But, if we add the above, it works fine.

ok, thanks for your comment~