QDECR 0.9.0

Version 0.9.0 is the first update after publication of the QDECR manuscript in Frontiers in Neuroinformatics. The latter is based in Lausanne, Switzerland. This version introduces weighted regression and a number of minor tweaks.

New features

• #47: qdecr_fastlm can now take weights via the weights argument, which act as observation weights in the linear regression. This argument works similar to lm (addresses #36).

Bug fixes

• #43: The specified number of cores is checked against bigstatsr::nb_cores (inside QDECR:::check_cores). However, it seems that it returns 0 cores when only 1 core is found (given that it always omits 1 core). We thus rewrote check_cores to use parallel::detectCores, and to make sure that it cannot return 0 cores (fixes #42).
• #43: Fixed the text in the error message of QDECR:::check_cores.
• #45: Unsmoothed q-cached surface files can now be analyzed. Normally, QDECR would look for files containing fwhmX, where X is the FWHM in mm. However, unsmoothed files do not contain this part. The code was rewritten to check if fwhm == 0, and in those cases it will not insert that into the file names. This was fixed by setting fwhmc to "" in QDECR:::qdecr_check, and by tweaking QDECR:::qdecr_prep_mgh (fixes #41).
• #51: Added an extra check to avoid multiple levels of parallelism, e.g. n_cores > 1 while simultaneously having a parallel BLAS library set up.
• 40f04d4: Added some functions to the NAMESPACE (stats::na.fail, stats::model.weights).
• 8f19d58: Since 4.0.0 it is possible to get a warning message from readChar. The load.annot function was modified by replacing readChar with the equivalent call of readBin.

Minor tweaks

• #46: Tweaked the vertex-wise analysis code (QDECR:::analysis_chunkedlm) to be faster and be more memory efficient (se is now calculated without the intermediate storage of s2) (addresses #38).
• 306826c: Added an extra check in QDECR:::qdecr_prep_mgh so that it also checks whether the .mgh files actually exist. It will output which subjects are missing the surface files (if N < 20) or just say that people are missing surface files (if N >= 20). This adds a little bit of runtime, but it pays off (fixes #34).
• 26ac549: When there is missing data in the design matrix, QDECR:::qdecr_fastlm will throw an error (i.e., na.action = na.fail). This is to avoid problems downstream. We opted for na.fail and not na.omit-like behavior, because users who are unaware of missingness would then only find out after running QDECR that they had missingness.