Add random seed for ANTs

[14thibea]

Hello all :)

For a project using ANTs we would need to have a deterministic setup. This is the aim of this PR in which a value can be given for the random seed of ANTs, thus leading to the same result (using the same computer).
Before merging it would be worth testing if the results are the same on different computers as well. Also this could be checked in the CI.

Best,

[ghisvail]

@14thibea shall I merge this as is, or are there other things you need to test? Your PR is still at draft stage atm.

[14thibea]

Hi @ghisvail
I am going to run t1-linear in another machine (Linux) to check I have the same result!
If it works I will try to put this on the CI as well :)

[14thibea]

I checked and I don't get exactly the same result on 2 different machines :(
But at least they are much closer than two different random seeds on the same machine...

[sophieloiz]

Hi @ghisvail, from what I understand, this PR addresses the random seed issue just for the t1-linear pipeline. Would it be possible to also add this seed for the flair-linear pipeline ?

[ghisvail]

Hi @ghisvail, from what I understand, this PR addresses the random seed issue just for the t1-linear pipeline.

Actually, both t1-linear and flair-linear use the same anat-linear pipeline underneath. So flair-linear will also benefit from the pieces of improvement proposed by @14thibea 👍

[sophieloiz]

Perfect, thanks ! Do we need to modify the flair_linear_cli.py file to add the --random_seed parameter ?

[ghisvail]

Just a quick heads-up:

• Following Friday's meeting, I added random seed support to pet-linear
• I took the liberty to do some refactoring in the process, i.e.:
  • Move the random seed option to cli_param as it is now used by more than one pipeline
  • Use assignment expressions since we are on Python 3.8+ and they simplify the code