Add parameter to allow equal weight during HPAL level 1 iteration. #532
Conversation
1 similar comment
Codecov Report
@@ Coverage Diff @@
## master #532 +/- ##
==========================================
- Coverage 76.66% 75.63% -1.04%
==========================================
Files 364 363 -1
Lines 41341 41370 +29
Branches 6667 6672 +5
==========================================
- Hits 31695 31289 -406
- Misses 7751 8231 +480
+ Partials 1895 1850 -45
Continue to review full report at Codecov.
|
|
Could you add some unit test code to test the new feature please? |
| ha = Hyperalignment(ref_ds=ref_ds, | ||
| zscore_all=zscore_all, | ||
| zscore_common=zscore_common) | ||
| zscore_common=zscore_common, | ||
| level1_equal_weight=level1_equal_weight) |
There was a problem hiding this comment.
that is a start! ;) although too expensive one I must say, but how do we know that the "logic" is actually working? ;-)
@feilong could you please add a simple test for mean_xy which would demonstrate that if we provide similar to the ones used in the code weighting, it would produce the desired equivalently balanced contribution?
I know that it sounds "not that interesting", but that is why they are called "unit-tests".
There was a problem hiding this comment.
No problem, just added a new commit with the tests for mean_xy.
|
|
| @@ -430,7 +437,12 @@ def _level1(self, datasets, commonspace, ref_ds, mappers, residuals): | |||
| # to make a batch update after processing all 1st-level datasets | |||
| # to an identical 1st-level common space | |||
| # TODO: make just a function so we dont' waste space | |||
| commonspace = params.combiner1(ds_, commonspace) | |||
| if params.level1_equal_weight: | |||
There was a problem hiding this comment.
who knows... probably still is (for "inplace" zscoring, right @swaroopgj?), but it is unrelated to this PR
yeap, not your fault... for now just restarted that run, let's pretend it didn't happen for now ;-) |
|
Seems to be LGTM, will merge |
I raised an issue, let's keep an eye on it. |
If this looks good to you I'll also update
searchlight_hyperalignment.pyto make similar changes to SL HA. @yarikoptic @swaroopgj