TransformedDistribution does not handle sample_shapes with >1 dimension

[feynmanliang]

d, param = flowtorch.bijectors.AffineAutoregressive(
    flowtorch.params.DenseAutoregressive()
)(dist.Independent(dist.Normal(torch.zeros(3), torch.ones(3)), 1))
d.rsample((10,2)).shape

should return the same [10,2,3] shape as dist.Independent(dist.Normal(torch.zeros(3), torch.ones(3)), 1).rsample((10,2)).shape, but it currently RuntimeErrors due to mismatched shape.

[stefanwebb]

Ah apologies, these shape tests are long overdue! I'll prioritize the docs for shapes, constraints, and bijectors, and make the adjustments to the Bijector interface. Then we'll be ready to write the unit tests for shapes (and Jacobians, inverses, etc.)

[feynmanliang]

The behavior on log_prob's shape is also inconsistent with torch.distributions and may be due to the same RC, will open separate issue to track

[feynmanliang]

Idea for fix: take product over sample_shape and then reshape