Graphical-Normalizing-Flows icon indicating copy to clipboard operation
Graphical-Normalizing-Flows copied to clipboard
verified publisher icon AWehenkel

How to calculate the inverse transform of Graphical Normalizing Flow?

Open EternityZY opened this issue 3 years ago • 2 comments

Hello, this is a great job There is a problem, in the experiment Conditioner=DAG, Normalizer=Monotonic How to calculate the inverse transformation of Graphical Normalizing Flow? That is, how to get the generated data X' from the random noise Z by using the inverse transform X = g^-1(Z). If this part of the code has been published, please tell me where it is, thank you

EternityZY avatar Jul 22 '22 16:07 EternityZY

Hi! I do not have my computer with me at the moment but I will take a look at this beginning of August. Inversion should work, it is based on the same algorithms as for the affine transformation except that each monotonic transformation is inverted with a binary search.

AWehenkel avatar Jul 23 '22 09:07 AWehenkel

Hi @EternityZY! The code is already online, can you explain to me what exactly you are not able to do?

AWehenkel avatar Aug 06 '22 14:08 AWehenkel