Features/lshape map+redistribute #428
Conversation
Codecov Report
@@ Coverage Diff @@
## master #428 +/- ##
==========================================
+ Coverage 96.58% 96.59% +0.01%
==========================================
Files 57 57
Lines 11981 12053 +72
==========================================
+ Hits 11572 11643 +71
- Misses 409 410 +1
Continue to review full report at Codecov.
|
| Returns | ||
| ------- | ||
| None, the local shapes of the DNDarray are modified | ||
| """ |
There was a problem hiding this comment.
a couple examples would be good, esp. about "the only important parts of the target map are the values along the split axis".
| if lshape_map.shape != (self.comm.size, len(self.gshape)): | ||
| raise ValueError( | ||
| "lshape_map must have the dimensions ({}, {}), currently {}".format( | ||
| self.comm.size, len(self.gshape), lshape_map.shape |
There was a problem hiding this comment.
Mix up between "dimensions" and "shape"? (lshape_map must have shape blah blah, currently blah)
| if not isinstance(target_map, torch.Tensor): | ||
| raise TypeError( | ||
| "target_map must be a torch.Tensor, currently {}".format(type(lshape_map)) | ||
| ) |
There was a problem hiding this comment.
format(type(lshape_map)) should be format(type(target_map))
| if target_map.shape != (self.comm.size, len(self.gshape)): | ||
| raise ValueError( | ||
| "target_map must have the dimensions {}, currently {}".format( | ||
| (self.comm.size, len(self.gshape)), target_map.shape |
There was a problem hiding this comment.
dimensions vs. shape, see above
Description
The balance function now uses a redistribute function. The logic is the same as it was before, but it can now be given an arbitrary shape to match.
Fixes: None, updates balance function in view of purposely unbalancing tensors for other operations
Changes proposed:
create_lshape_mapfunction to generate the lshape of a functionType of change
Select relevant options.
Are all split configurations tested and accounted for?
[x] yes [ ] no
Does this change require a documentation update outside of the changes proposed?
[ ] yes [x] no
Does this change modify the behaviour of other functions?
[x] yes [ ] no
Are there code practices which require justification?
[ ] yes [x] no