First Level Bayesian Estimation fails

[gladomat]

Summary

The Bayesian estimation isn't fully implemented in nipype.

Actual behavior

Since the Bayesian implenetation writes Cbeta imags instead of beta images, the code runs into problems. For example it cannot find an entry in the SPM.mat with the name Vbeta, and rightly so, because this is named VCbeta. This is unfortunate, as I had to wait 4 days for this error to appear (since Bayesian estimation takes ~10 times longer).

Traceback (most recent call last):
  File "//nobackup/archimedes1/Glad/anaconda2_serverwide/lib/python2.7/site-packages/nipype/pipeline/plugins/multiproc.py", line 49, in run_node
    result['result'] = node.run(updatehash=updatehash)
  File "//nobackup/archimedes1/Glad/anaconda2_serverwide/lib/python2.7/site-packages/nipype/pipeline/engine/nodes.py", line 394, in run
    self._run_interface()
  File "//nobackup/archimedes1/Glad/anaconda2_serverwide/lib/python2.7/site-packages/nipype/pipeline/engine/nodes.py", line 504, in _run_interface
    self._result = self._run_command(execute)
  File "//nobackup/archimedes1/Glad/anaconda2_serverwide/lib/python2.7/site-packages/nipype/pipeline/engine/nodes.py", line 630, in _run_command
    result = self._interface.run()
  File "//nobackup/archimedes1/Glad/anaconda2_serverwide/lib/python2.7/site-packages/nipype/interfaces/base.py", line 1044, in run
    outputs = self.aggregate_outputs(runtime)
  File "//nobackup/archimedes1/Glad/anaconda2_serverwide/lib/python2.7/site-packages/nipype/interfaces/base.py", line 1115, in aggregate_outputs
    predicted_outputs = self._list_outputs()
  File "//nobackup/archimedes1/Glad/anaconda2_serverwide/lib/python2.7/site-packages/nipype/interfaces/spm/model.py", line 228, in _list_outputs
    for vbeta in spm['SPM'][0, 0].Vbeta[0]:
AttributeError: 'mat_struct' object has no attribute 'Vbeta'
Interface EstimateModel failed to run. 

Furthermore, I think the contrast estimate implementation will also fail, since no beta images are written, but instead Cbeta_###.nii are written. Nipype won't look for these either.

Expected behavior

There should be a check if Bayesian estimation is selected, in order to select VCbeta instead of Vbeta in /spm/model.py. Downstream processing should also account for the fact that Cbeta images are written (instead of beta*.nii). This ensures contrast estimation calculation using posterior probability maps (PPMs).

How to replicate the behavior

Run a first level Bayesian estimation with a really small mask (1mm sphere) in order to speed up the analysis.

Platform details:

I'm getting this error but I'm using nipype-0.12.1-py27_0:

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/nobackup/archimedes1/Glad/anaconda2_serverwide/lib/python2.7/site-packages/nipype/__init__.py", line 60, in <lambda>
    get_info = lambda: _get_pkg_info(os.path.dirname(__file__))
  File "/nobackup/archimedes1/Glad/anaconda2_serverwide/lib/python2.7/site-packages/nipype/pkg_info.py", line 82, in get_pkg_info
    src, hsh = pkg_commit_hash(pkg_path)
  File "/nobackup/archimedes1/Glad/anaconda2_serverwide/lib/python2.7/site-packages/nipype/pkg_info.py", line 52, in pkg_commit_hash
    archive_subst = cfg_parser.get('commit hash', 'archive_subst_hash')
  File "/home/raid2/mihai/.local/lib/python2.7/site-packages/backports/configparser/__init__.py", line 811, in get
    d)
  File "/home/raid2/mihai/.local/lib/python2.7/site-packages/backports/configparser/__init__.py", line 384, in before_get
    self._interpolate_some(parser, option, L, value, section, defaults, 1)
  File "/home/raid2/mihai/.local/lib/python2.7/site-packages/backports/configparser/__init__.py", line 434, in _interpolate_some
    "found: %r" % (rest,))
backports.configparser.InterpolationSyntaxError: '%' must be followed by '%' or '(', found: u'%h'

[satra]

@gladomat - thanks for reporting this. we will take a look.

[mgxd]

@gladomat thanks for this, as a side note do you know if Bayesian2 estimation also write these Cbeta images?

[mgxd]

@gladomat and as for your error, try pip install configparser

[gladomat]

I'm on vacation right now, and I can't remember. The second level Bayesian first requires a second level frequentist or classical inference. Here the betas get written. You then have to re-estimate the whole thing using Bayesian stats, by loading the estimated spm.mat and running the estimation step. You can try it with first level classical inference data; just run a second level Bayesian estimation on the spm.mat file. Am 30. März 2017 03:29:31 GMT+09:00 schrieb Mathias Goncalves <notifications@github.com>:

…

@gladomat thanks for this, as a side note do you know if `Bayesian2` estimation also write these `Cbeta` images? -- You are receiving this because you were mentioned. Reply to this email directly or view it on GitHub: #1786 (comment)

-- Diese Nachricht wurde von meinem Android-Gerät mit K-9 Mail gesendet.

[gladomat]

@mgxd The second level Bayesian estimation writes first a beta image for the classical estimation, then writes a Cbeta for the Bayesian estimation.

[mgxd]

@gladomat thanks for the information - I've created a branch that should hopefully solve this. It would be great if you could try it out and let me know!

[gladomat]

@mgxd I definitely will, but it'll have to wait a bait. I hope that's ok.

[mgxd]

@gladomat no problem, i'll hold off submitting a PR until it's tested

[mgxd]

@gladomat this should now be fixed after #2030 is merged

[gladomat]

I'm sorry, I haven't gotten around to it. Does this mean you won't wait for my testing? This is of course fine; I'm just asking.

[mgxd]

I ended up testing the interface during another fix :)