Implement Symmetric and Asymmetric Multivariate Laplace distributions #389
Conversation
ricardoV94
force-pushed
the
mv_laplace
branch
2 times, most recently
from
c98ff2e to
b5e1b41
Compare
|
|
||
|
|
||
| @pytest.mark.xfail(reason="Not sure about equivalence. Test fails") | ||
| def test_asymmetric_matches_univariate_logp(): |
There was a problem hiding this comment.
I was expecting these to be equivalent. The means and variances match, but not the logp. Either I did something wrong or the equivalence is not real / like this.
| from pytensor.tensor.random.utils import normalize_size_param | ||
|
|
||
|
|
||
| class Kv(BinaryScalarOp): |
There was a problem hiding this comment.
I'll move this to PyTensor before merging this PR
ricardoV94
force-pushed
the
mv_laplace
branch
from
b5e1b41 to
e3aab90
Compare
ricardoV94
force-pushed
the
mv_laplace
branch
from
e3aab90 to
e1680f0
Compare
|
Deprecation failures should be fixed by pymc-devs/pymc#7564 |
|
@ricardoV94 I think this needs to be rebased. There are references to |
|
Yes it does, but I won't bother if I get no reviews haha |
| [out] = outputs | ||
| [g_out] = output_grads | ||
| dx = -(v / x) * out - self.kv(v - 1, x) | ||
| return [grad_not_implemented(self, 0, v), g_out * dx] |
There was a problem hiding this comment.
Does this work without a grad on this Op?
| _print_name = ("MultivariateAsymmetricLaplace", "\\operatorname{MultivariateAsymmetricLaplace}") | ||
|
|
||
| @classmethod | ||
| def rv_op(cls, mu, cov, *, size=None, rng=None): |
There was a problem hiding this comment.
The rv_op for MvAsymmetricLaplaceRV and MvLaplaceRV seem almost identical. Is it worth abstracting that?
| return super().dist([mu, cov], **kwargs) | ||
|
|
||
|
|
||
| @_logprob.register(MvLaplaceRV) |
There was a problem hiding this comment.
Is there a reason to prefer @_logprob.register over a logp method? Or is one just the new way to do this now?
TODO: Bump PyMC dependency