Figure out how to test multiple modelarray versions

[mattcieslak]

Maybe we can pull tagged versions of ModelArray from Dockerhub
Then run an Rscript

Sub Issues

• can Dependabot figure out when there is a new release of ModelArray?
• We might be able to have a checker run on a cron job and update a compatibility chart (custom dependabot)
• Should it only do Major releases?
• We should hardcode a specific R version: None of the dependency packages will impact the indexing of the input/output data
• We need to check if there are forward/backward incompatibilities in HDF5 and tiledb. These could be important

Maybe test a version matrix?
https://stackoverflow.com/questions/40179493/r-how-can-i-install-a-specific-release-by-install-github

[tsalo]

Scheduled job workflow:

1. Use GitHub API to build list of releases of ModelArray.
2. Filter list of releases to only include first release in each major release series.
3. Filter list of releases to only include releases after or equal to minimum ModelArray version supported by current ModelArrayIO (have to hardcode this somewhere).
4. Build version matrix from list of ModelArray releases.
   • Since we're only testing mean, which is base R, we don't need to test across R versions.
5. Run ModelArrayIO on test data, pass hd5/tiledb files to ModelArray, compute mean, pass back to ModelArrayIO, then compare against test ground truth files.

Test data creation:

• We want NIfTI, CIFTI, and fixel files (10 single-volume files) where each voxel's value is an integer and each mean value across files is equal to the current index of that voxel. The ground truth files will be the current index for each voxel in that file. The exact order of indices doesn't matter as long as the values are unique and can be compared to ground truth.