Open
Description
Is there a way to implement a tfp.monte_carlo.expectation passing N samples for multiple random variables with different dimensions?
E.g.
tfp.monte_carlo.expectation( f=lambda s: ...,
samples = samples,
...)
where samples are comprised from two tensors with shapes (N, 2) and (N,2,2) ? Is that possible by defining samples as a list of N tuples of tensors constructed on slices of the two tensors along their first dimension?