probability
probability copied to clipboard
Glow coupling blocks conditioning
Hi, I am currently using the tensorflow probability Glow model (tfb.Glow) and I am interested on conditioning the coupling blocks on a different tensor than the input data. Can anyone provide directions on how to properly implement it?
Thanks in advance
There's no easy way to do it at the moment. The following enhancements are necessary:
- The two
bijector_fnfunctions (1 and 2) need to accept and pass along conditioning kwargs. - All the bijectors inside glow need to have unique names, so we can target them using code similar to this snippet.
Alternatively, you could always just capture your auxiliary input into the callables you pass to construct your glow.