Optional wm

[nicholst]

Collection of changes to make use of WM mask optional and settable from command line, and have mask name be set from command line, and uniformly use .nii.gz.

[nicholst]

@bathybicBee : I've done it... I've beaten Tim's code into submission... or well, tried to tame it some. This PR has a number of changes for which I'd like you to update the documentation:

• Usage has changed

  Usage: BinCar  NTypes  NCov  GPU  Design Mask WM [MaxIter BurnIn]  
  NTypes  - Number of groups
  NCov    - Number of covariates (count must include groups)
  GPU     - 1 use GPU; 0 use CPU (CPU not tested! Use with caution)
  Design  - Text file, tab or space separated data file
  Mask    - Filename of mask image in './images' directory, or '1' to use default (mask.nii.gz)
  WM      - Filename of WM image in './images' directory, '0' to use none, or '1' to use default (avg152T1_white_.nii.gz)
  MaxIter - Number of iterations (defaults to 1,000,000)
  BurnIn  - Number of burn-in iterations (defaults to 500,000)
  For documentation see: http://warwick.ac.uk/tenichols/BSGLMM
  

  Notably, mask and WM now must be specified, but they have short-cut values (1 for mask, 0 or 1 for WM).

• All images now must be .nii.gz. Before (oddly, and annoying for the code) the white matter image was {hdr,img}.
• When WM is 0, the beta is not updated, as we were doing manually by commenting out that section of code.

These changes were facilitated by new, general code in read_data.cpp for interacting with NIFTI files.

Once you've documented these changes I'll make a new release and send to Basel, though, frankly, the changes are mainly to make it easier to use in non-MS settings.

Now the only remaining serious changes I'd like to make are

1. Allowing the file type to be either nii or nii.gz, and letting the user specify either (or, even, no extension) and having the code work out which to use.
2. Dropping the necessity of having all images in ./images

[btaschler]

@nicholst : I have updated the documentation on the website to reflect the new usage.

Some minor issues about the produced output:

• Do we want to keep the diagnostic output at every 100 iterations ?
• Some of the additional output files were removed for the initial release:
  • bWM.nii - proportional to WM, i.e. coefficients times WM mask
  • DIC.dat - storing the following: deviance of the expectation (DE), expectation of deviance (ED), effective degrees of freedom (PD), deviance information criterion (DIC)
• Other output files (also not included initially) are empty:
  • fBF.dat
  • parms.dat
  • select_parms.dat
  • resid.dat

[nicholst]

Yes, keep the diagnostic output. It is useful and cheap.

bWM.nii should not be written if UPDATE_WM == 0.

If The DIC info is valid (and I assume it is) it should be retained.

Please eliminate any/all output files that are always empty.

Not sure what's the best way to proceed. Maybe merge my pull request, and then make a new PR with these edits. Or, if you know how, check out OptionalWM branch and make a PR upon that branch.

@nicholst : I have updated the documentation on the website to reflect the new usage.

Some minor issues about the produced output:

Do we want to keep the diagnostic output at every 100 iterations ?

Some of the additional output files were removed for the initial release:

bWM.nii - proportional to WM, i.e. coefficients times WM mask DIC.dat - storing the following: deviance of the expectation (DE), expectation of deviance (ED), effective degrees of freedom (PD), deviance information criterion (DIC) Other output files are empty:

fBF.dat parms.dat select_parms.dat resid.dat — Reply to this email directly or view it on GitHub.