Release SnPM 13.1.08

[cmaumet]

This PR includes updates for the next release of SnPM:

• fix: fix error where large ANOVA-within-subject would fail with "SnPM:InvalidPiCond" (#61, https://github.com/cmaumet/SnPM-devel/pull/1)
• fix: dimensions when 'low mem' cluster inference done in 2D (#69)
• enh: Table formatting (#51)
• enh: Updated cluster forming threshold default to 2.3 (#59)
• enh: allow for covariates in ANOVA within- and between-subject (#61, #60)
• enh: Adding Zenodo's badge to the README (with doi) (#62)
• dev: Add template for pull requests (#72)

[cmaumet]

Hi @nicholst,

Tests currently report 1 failure:

Failure Summary:

     Name                                                              Failed  Incomplete  Reason(s)
    =================================================================================================
     test_multisub_withinsubanova/test_multisub_withinsubanova_approx    X         X       Errored.

Totals:
   69 Passed, 1 Failed, 1 Incomplete.
   525.0753 seconds testing time.

It looks like changes in https://github.com/SnPM-toolbox/SnPM-devel/pull/61, in particular https://github.com/SnPM-toolbox/SnPM-devel/pull/61/commits/e38507307c12ee29cbfe98eed9a498c44b463b77, are breaking the tests. Complete log:

>> run(test_multisub_withinsubanova, 'test_multisub_withinsubanova_approx')
Running test_multisub_withinsubanova

------------------------------------------------------------------------
Running job #6
------------------------------------------------------------------------
Running 'Within Subject ANOVA, k diffs/contrasts per subject'
Failed  'Within Subject ANOVA, k diffs/contrasts per subject'
Error using snpm_pi_ANOVAwithinS (line 280)
Bad PiCond ()
In file "/Users/camaumet/Softs/SnPM-devel/spm12/toolbox/SnPM13/snpm_pi_ANOVAwithinS.m" (???), function "snpm_pi_ANOVAwithinS" at line 280.
In file "/Users/camaumet/Softs/SnPM-devel/spm12/toolbox/SnPM13/snpm_ui.m" (???), function "snpm_ui" at line 297.
In file "/Users/camaumet/Softs/SnPM-devel/spm12/toolbox/SnPM13/config/snpm_run_ui.m" (???), function "snpm_run_ui" at line 12.

No executable modules, but still unresolved dependencies or incomplete module inputs.
The following modules did not run:
Failed: Within Subject ANOVA, k diffs/contrasts per subject
Skipped: Compute
Skipped: Inference

@nicholst: can you have a look to propose a fix? I would change ~(nSubj<=12 || ~bAproxTst) into (nSubj<=12 || ~bAproxTst) but not sure whether that would break the case you were trying to fix in #61.

[nicholst]

Ug... this is annoying. I don't think the fix you suggest works. The setup is that we have 'exact' and 'random' methods...

if nSubj<=12 || ~bAproxTst                    % exact method
 [...]    
else                                          % random method
 [...]
end

and then

elseif length(perm)==0 && ~(nSubj<=12 || ~bAproxTst)
    % Special case where random method missed iCond; order of perms is
    % random so can we can just replace first perm.
    PiCond(1,:) = iCond;
    perm = 1;

So the logic seems right... if the correctly labeled permutation isn't found AND NOT(exact) then you can add it to the top; but if the correctly labeled permutation isn't found AND Exact there's a problem.

Can you direct me to the test case that generates this? I'll have a look... though realistically not before Thursday.

[cmaumet]

Right! I always get these wrong...

Anyway, the failing test case is test_multisub_withinsubanova_approx:

• code is here
• corresponding ground truth test data (including SnPMcfg.mat) is here.

Let me know!

[nicholst]

OK @cmaumet! Hopefully https://github.com/cmaumet/SnPM-devel/pull/1 will fix this... hallelujah for testing!

[cmaumet]

Thanks @nicholst! Your update fixed the issue but then the tests were failing because the fix also introduced a change in the selected permutations for approximate analyses.

This is a problem for the non-regression tests were we check that the produced outputs are identical to the ground truth outputs. We could update SnPM8 and recompute the ground truth to match what you have done in https://github.com/SnPM-toolbox/SnPM-devel/pull/70/commits/264fb2c7ea1bea5d1c3b5e2668ae87d558c48d6b. Another solution is to change slightly your update to preserve the order we had before for the permutations. That's what I've done in 12872ece4c0f4b2594ee5a958a29e57b4a549fa6. Let me know if that's fine with you.

All tests now pass:

Totals:
   70 Passed, 0 Failed, 0 Incomplete.
   179.9142 seconds testing time.

[nicholst]

I thought of this belatedly.

Yes, I'm fine with this. Thank you!