FIX: FEMA input index should filter on rows, not columns.

[adelavega]

Unfortunately, I didn't test this thoroughly enough, and the update on the FEMA input file indexing was incorrect. I reverted to a more clear/optimized version of my original logic.

Consider the following:

filtered_effects = ['/tmp/tmpe9vfsb1a/fitlins_wf/l1_model/mapflow/_l1_model0/text_effect_size.nii.gz',                                                                                                                         
 '/tmp/tmpe9vfsb1a/fitlins_wf/l1_model/mapflow/_l1_model0/rmse_effect_size.nii.gz',                                                                                                                         
 '/tmp/tmpe9vfsb1a/fitlins_wf/l1_model/mapflow/_l1_model0/speech_effect_size.nii.gz',                                                                                                                       
 '/tmp/tmpe9vfsb1a/fitlins_wf/l1_model/mapflow/_l1_model1/text_effect_size.nii.gz',                                                                                                                         
 '/tmp/tmpe9vfsb1a/fitlins_wf/l1_model/mapflow/_l1_model1/rmse_effect_size.nii.gz',                                                                                                                         
 '/tmp/tmpe9vfsb1a/fitlins_wf/l1_model/mapflow/_l1_model1/speech_effect_size.nii.gz',                                                                                                                       
 '/tmp/tmpe9vfsb1a/fitlins_wf/l1_model/mapflow/_l1_model2/text_effect_size.nii.gz',                                                                                                                         
 '/tmp/tmpe9vfsb1a/fitlins_wf/l1_model/mapflow/_l1_model2/rmse_effect_size.nii.gz',                                                                                                                         
 '/tmp/tmpe9vfsb1a/fitlins_wf/l1_model/mapflow/_l1_model2/speech_effect_size.nii.gz']  

# After running `prepare_contrasts`
contrasts = [('rmse', array([[1, 0, 0]]), 'FEMA'),                                                                                                                                                                      
 ('speech+rmse', array([[1, 0, 1]]), 'FEMA')]

and the design matrix looks like:

In [26]: mat                                                                                                                                                                                                
Out[26]:                                                                                                                                                                                                    
   rmse  speech  text                                                                                                                                                                                       
0     0       0     1                                                                                                                                                                                       
1     1       0     0                                                                                                                                                                                       
2     0       1     0                                                                                                                                                                                       
3     0       0     1                                                                                                                                                                                       
4     1       0     0                                                                                                                                                                                       
5     0       1     0                                                                                                                                                                                       
6     0       0     1                                                                                                                                                                                       
7     1       0     0                                                                                                                                                                                       
8     0       1     0  

For the first contrast [1, 0, 0] , you want the files indicated by 1s in the rmse column.

The current logic of ix = weights[0].astype(bool) yields:

Out[83]: array([ True, False, False])

which is not what we want, because it's not the correct shape for filtering filtered_effects.

I think this slipped past me because in my test analysis it just so happened that filtered_effects was of length 3, and it did not crash (although it probably filtered the wrong things).

Instead, we want to use that ix to filter mat.

con_ix = weights[0].astype(bool)
dm_ix = mat.iloc[:, con_ix].any(axis=1)

In [87]: dm_ix
Out[87]: 
0    False
1     True
2    False
3    False
4     True
5    False
6    False
7     True
8    False
dtype: bool

Now the reason that I added the any(axis=1), is because of second contrast with weights [1, 0, 1].

In this case we want to filter mat to select those two columns, and then select any rows that have a non-zero value in either columns:

In [98]: dm_ix
Out[98]: 
0     True
1     True
2    False
3     True
4     True
5    False
6     True
7     True
8    False
dtype: bool

Now, this might be a weird FEMA to specify, but it's theoretically valid.

[adelavega]

As an aside, this is obviously means that FEMA is still broken in 0.6.1. Since I'm the only user of this, I'd be happy to just pin fitlins to this commit, and build from 0.6.1. Or we can do another hotfix release, your call.

[effigies]

Looks reasonable. I appreciate the example to walk through, because thinking through these things is painful.

I'm good with going ahead and doing a 0.6.2 release.

[codecov-io]

Codecov Report

Merging #207 into master will increase coverage by 0.02%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #207      +/-   ##
==========================================
+ Coverage   77.35%   77.37%   +0.02%     
==========================================
  Files          18       18              
  Lines        1042     1043       +1     
  Branches      187      187              
==========================================
+ Hits          806      807       +1     
- Misses        147      148       +1     
+ Partials       89       88       -1

Flag	Coverage Δ
#ds003	77.37% <100%> (+0.02%)	⬆️

Impacted Files	Coverage Δ
fitlins/interfaces/nistats.py	83% <100%> (+0.11%)	⬆️
fitlins/interfaces/bids.py	73.33% <0%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a53f76e...86aec54. Read the comment docs.