Support logcdf, logccdf and icdf for dim RVs #8083
Conversation
ricardoV94
force-pushed
the
support_more_logp_methods_dims
branch
2 times, most recently
from
6464254 to
f1b70ef
Compare
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #8083 +/- ##
=======================================
Coverage 90.89% 90.90%
=======================================
Files 123 123
Lines 19489 19506 +17
=======================================
+ Hits 17714 17731 +17
Misses 1775 1775
🚀 New features to boost your workflow:
|
ricardoV94
force-pushed
the
support_more_logp_methods_dims
branch
from
f1b70ef to
d1015d9
Compare
| @_logprob.register(MeasurableXTensorFromTensor) | ||
| def measurable_xtensor_from_tensor(op, values, rv, **kwargs): | ||
| rv_logp = _logprob(rv.owner.op, tuple(v.values for v in values), *rv.owner.inputs, **kwargs) | ||
| def _to_xtensor(expr, op: MeasurableXTensorFromTensor): |
There was a problem hiding this comment.
nit: expr is a vague variable name. I associated it with polars expressions, definitely would not have guessed that we want to pass one of _logprob, _logcdf or _icdf here
| x = Normal.dist( | ||
| mu=as_xtensor(mean, dims=("city",)), | ||
| sigma=1, | ||
| dim_lengths={"time": 5}, |
There was a problem hiding this comment.
There's no time dim declared, is this a non-public API for making a dims-aware RV?
There was a problem hiding this comment.
This is how you specify shape in pymc.dims/xtensor.random variables. There's no size/shape argument, only this. When you are in a model context with a named variable, it forwards this information from model.dim_lengths, but if you are working with a .dist you work with this manually.
|
Approved but there's a failing test, i guess unrelated. |
2 participants
Only
_logprobhad a dispatch before