FIX: Handle correct contrasts and intercepts being passed from PyBIDS #387
Conversation
Codecov ReportBase: 64.45% // Head: 64.18% // Decreases project coverage by
Additional details and impacted files@@ Coverage Diff @@
## dev #387 +/- ##
==========================================
- Coverage 64.45% 64.18% -0.27%
==========================================
Files 23 23
Lines 1820 1829 +9
Branches 345 298 -47
==========================================
+ Hits 1173 1174 +1
- Misses 565 570 +5
- Partials 82 85 +3
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here. ☔ View full report at Codecov. |
|
@adelavega @jmumford @jdkent Looks like 0.15.3 runs, but now we're getting different values in our results. Suggests we're now using different design matrices, as nilearn and AFNI aren't changing, and neither are the input masks. Do any of you remember changes that we would have expected to improve design matrix construction? |
|
Could it be a change in the smoothing or highpass filtering options? |
|
Did we change those in PyBIDS? The only difference between this and the failing tests on the dev branch is that this is using pybids 0.15.1 and not 0.15.3. |
|
Hmm. I guess I'd like to see a diff between the two DMs. The only thing I can think of is that we were not modeling an intercept is some situations, and possibly we are adding an intercept where we shouldn't be? Unless there is a major bug (i.e. dropped columns), not sure where else the DMs would differ |
|
@adelavega Yes, the difference is that now there's an explicit |
|
FWIW I did write up my method for reproducing, since I wanted to get it set up on another system. https://github.com/poldracklab/fitlins/wiki/Testing-FitLins-regressions-on-example-data |
|
Awesome, so that should be a fitlins issue then, correct? |
53e2c57 to
2ed5e36
Compare
a403640 to
6bb846f
Compare
|
Well, it fixes the regression test, so that's good. Might have revealed other bugs in FitLins, though... |
No description provided.