include quantile function in distribution #314
Conversation
|
Job PR-314/1 is complete. |
vafl
force-pushed
the
quantile_function
branch
from
2f1f193
to
c91bd79
Compare
|
Job PR-314/2 is complete. |
Codecov Report
@@ Coverage Diff @@
## master #314 +/- ##
===========================================
+ Coverage 40.45% 80.54% +40.08%
===========================================
Files 142 142
Lines 7966 8024 +58
===========================================
+ Hits 3223 6463 +3240
+ Misses 4743 1561 -3182
|
| @@ -223,6 +230,12 @@ def cdf(self, x: Tensor) -> Tensor: | |||
| """ | |||
| raise NotImplementedError() | |||
|
|
|||
| def quantile(self, level: Tensor) -> Tensor: | |||
| r""" | |||
| Returns the quantile corresponding to the passed level | |||
There was a problem hiding this comment.
It would be better to document level here as well as the return value (shape and so on)
| def quantile(self, level: Tensor) -> Tensor: | ||
| F = self.F | ||
| # we consider level to be an independent axis and so expand it | ||
| # to shape (num_levels, 1, 1, ...) |
There was a problem hiding this comment.
For example, this expansion here is a bit counterintuitive if one reasons in terms of numpy’s broadcasting
There was a problem hiding this comment.
But yeah, I guess we should be looking at how mxnet does broadcasting :-)
There was a problem hiding this comment.
Yeah, we may need to iterate a bit on this. For now I assume the simplest case where level is a 1d array and the method returns a tensor with (num_levels, ...DISTRIBUTION SHAPE...). I documented this in the base class.
vafl
force-pushed
the
quantile_function
branch
from
c91bd79
to
adcffaf
Compare
|
Job PR-314/3 is complete. |
Issue #, if available:
Description of changes:
Include a
quantilefunction for distributions.By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.