PennLINC / ModelArray

ModelArray: an R package for statistical analysis of fixel-wise data and beyond
https://pennlinc.github.io/ModelArray
BSD 3-Clause "New" or "Revised" License
5 stars 2 forks source link

[ENH] build docker image for ModelArray + ConFixel #44

Closed zhao-cy closed 2 years ago

zhao-cy commented 2 years ago
zhao-cy commented 2 years ago

Current docker image is from rocker/verse, it's not rocker/rstudio (https://hub.docker.com/r/rocker/rstudio), so it won't support rstudio or forward rstudio to a portal.

zhao-cy commented 2 years ago

confirmed: all run by singularity image from chenyingzhao/modelarray_confixel:test0.0.0, from h5 --> ModelArray --> nii.gz: replicated Kristin's PNC fMRI n-back results, with: the max absolute difference are minimal: For parentEdu: tstat and beta estimation: max abs diff < 5.367881747986303e-07. See PennLINC/ModelArray_tests: pncNback_readme.md

zhao-cy commented 2 years ago

when using circle ci to build docker image, probably want to specify --ncpus -4 to "--ncpus -2"; otherwise there may be error message from circle ci...

zhao-cy commented 2 years ago

ConFixel (fixel-part) depends on MRtrix's mrconvert. To save space, only need to copy the binary file mrconvert and eigen (C++ linear algebra dependencies)? ref: https://github.com/PennLINC/qsiprep_build/blob/main/Dockerfile L14: FROM pennbbl/qsiprep-mrtrix3:${TAG_MRTRIX3} as build_mrtrix3 to change to: FROM mrtrix/mrtrix3:${TAG_MRTRIX3} as build_mrtrix3 L55 (packages to be installed): MRTRIX3_DEPS="bzip2 ca-certificates curl libpng16-16 libblas3 liblapack3" I probably only need bzip2, and lib*