Feature/669 bincount #670
Feature/669 bincount #670
Conversation
Codecov Report
@@ Coverage Diff @@
## master #670 +/- ##
=======================================
Coverage 97.37% 97.38%
=======================================
Files 87 87
Lines 16711 16758 +47
=======================================
+ Hits 16273 16320 +47
Misses 438 438
Continue to review full report at Codecov.
|
|
rerun tests |
5 similar comments
|
rerun tests |
|
rerun tests |
|
rerun tests |
|
rerun tests |
|
rerun tests |
mtar
force-pushed
the
Feature/669-bincount
branch
from
68f081f to
6807be1
Compare
|
rerun tests |
1 similar comment
|
rerun tests |
heat/core/statistics.py
Outdated
| counts = torch.bincount(x._DNDarray__array, weights, minlength) | ||
|
|
||
| size = counts.size()[0] | ||
| maxlength = x.comm.allreduce(size, op=MPI.MAX) |
There was a problem hiding this comment.
Interesting. I would have just calculated maxlength = x.max()+1. You're bypassing all the __reduce_op() checks so you're probably more efficient :)
There was a problem hiding this comment.
I was curious. I get roughly 0.0001s ~ 0.0002s faster times in a very simple test.
heat/core/statistics.py
Outdated
| Raises | ||
| ------ | ||
| ValueError | ||
| If split axes of `x` and `weights` are not the same. |
There was a problem hiding this comment.
Technically this can't be because both only have one axis.
There was a problem hiding this comment.
It happens in the one case with None and 0, but pytorch would be complaining anyway because of different input tensor lengths.
There was a problem hiding this comment.
Yes, I understood what you were referring to, I was just suggesting to reword this line because None is not a split axis. Suggestions:
- "if the
splitattribute is not the same"? - "if
xis distributed andweightsisn't, or vice versa"?
|
rerun tests |
Description
Introduces the bincount function similar to the respective numpy and pytorch functions.
Issue/s resolved: #669
Changes proposed:
Type of change
Due Diligence
Does this change modify the behaviour of other functions? If so, which?
no