Fixing docker build for 1.0

[sarthakpati]

Fixes issue #(Enter issue number here)

Proposed Changes

• completely Docker-compatible build of the FeTS Tool 1.0
• Docker recipe is trying to replicate the linux installer script

[hasan7n]

@sarthakpati Error when running CLI Segment on sample data:

Traceback (most recent call last):
  File "/Front-End/bin/install/appdir/usr/bin//OpenFederatedLearning//submodules/fets_ai/Algorithms/fets/bin/brainmage_validation_scores_to_disk.py", line 20, in <module>
    from fets.models.pytorch.brainmage.brainmage import BrainMaGeModel
  File "/Front-End/bin/install/appdir/usr/bin/OpenFederatedLearning/venv/lib/python3.7/site-packages/fets/models/pytorch/brainmage/__init__.py", line 1, in <module>
    from .brainmage import BrainMaGeModel
  File "/Front-End/bin/install/appdir/usr/bin/OpenFederatedLearning/venv/lib/python3.7/site-packages/fets/models/pytorch/brainmage/brainmage.py", line 41, in <module>
    from openfl.models.pytorch import PyTorchFLModel
ModuleNotFoundError: No module named 'openfl.models.pytorch'
Using the following directory as logging directory: /data_renamed/logs
Starting subject directory iteration...
= Starting inference for subject: BraTS-GLI-00015-000
== Starting inference using FeTS Singlet Consensus model...
WARNING: The singlet model '52' did not run, please contact [email protected] with this error.
(I removed the rest as they are repetitive)

Hoping the following helps: inside the container, I saw that the following folder exists:

/Front-End/bin/install/appdir/usr/bin/OpenFederatedLearning/openfl/models/pytorch.

However, I saw that this

/Front-End/bin/install/appdir/usr/bin/OpenFederatedLearning/venv/lib/python3.7/site-packages/openfl/models

does not contain a pytorch folder. It contains a folder named dummy and three files: __init__.py, flmodel.py, and inference_only_model_wrapper.py
I wanted to look further but I think you are already more familiar. Thanks!

[hasan7n]

@sarthakpati
BTW I want to share as well every command I am running to build the docker image (in case I am missing something, plz let me know) and run the command

git clone --branch fixing_docker_build_for_1.0 https://github.com/FeTS-AI/Front-End
cd Front-End
git submodule sync
git -c protocol.version=2 submodule update --init --force --depth=1
docker build -t local/fets-tool:1.0.3 .

Then, inside the container, I am running:
FeTS_CLI_Segment -d /data_renamed -a fets_singlet,fets_triplet -lF STAPLE,ITKVoting,SIMPLE,MajorityVoting -g 0 -t 0

where data_renamed is some data I copied to the container:

root@8d1baf9a64c9:/Front-End# tree /data_renamed/
/data_renamed/
|-- BraTS-GLI-00001-001
|   |-- BraTS-GLI-00001-001_flair.nii.gz
|   |-- BraTS-GLI-00001-001_t1.nii.gz
|   |-- BraTS-GLI-00001-001_t1ce.nii.gz
|   `-- BraTS-GLI-00001-001_t2.nii.gz
`-- BraTS-GLI-00015-000
    |-- BraTS-GLI-00015-000_flair.nii.gz
    |-- BraTS-GLI-00015-000_t1.nii.gz
    |-- BraTS-GLI-00015-000_t1ce.nii.gz
    `-- BraTS-GLI-00015-000_t2.nii.gz

2 directories, 8 files

[sarthakpati]

/Front-End/bin/install/appdir/usr/bin/OpenFederatedLearning/venv/lib/python3.7/site-packages/openfl/models

This not containing the pytorch folder is confounding to me. By my estimate, this command [ref] should have ensured that the pytorch models in the Algorithms submodule would have gotten installed.

Perhaps @brandon-edwards can help?

[brandon-edwards]

I would have thought it would be here or the line below when the openfl.pytorch module would have been installed. I will stay with this discussion to help where I can, but I am not the one to get this resolved quickly. @msheller do you have any insight here?

[sarthakpati]

@hasan7n can you try with the latest commit?

[hasan7n]

@sarthakpati Thanks this resolves the problem.
Now I am getting this:

Using the following directory as logging directory: /data_renamed/logs
Starting subject directory iteration...
= Starting inference for subject: BraTS-GLI-00015-000
== Starting inference using FeTS Singlet Consensus model...
usage: brainmage_validation_scores_to_disk.py [-h] --data_path DATA_PATH
                                              --plan_path PLAN_PATH
                                              --model_weights_path_wt
                                              MODEL_WEIGHTS_PATH_WT
                                              --model_weights_path_et
                                              MODEL_WEIGHTS_PATH_ET
                                              --model_weights_path_tc
                                              MODEL_WEIGHTS_PATH_TC
                                              --output_pardir OUTPUT_PARDIR
                                              [--model_output_tag MODEL_OUTPUT_TAG]
                                              [--device DEVICE]
                                              [--process_training_data]
brainmage_validation_scores_to_disk.py: error: unrecognized arguments: -md cpu
WARNING: The singlet model '52' did not run, please contact [email protected] with this error.

[sarthakpati]

Thanks. For some reason, the image on my local machine is picking up a different version of script. Anyway, pushed a fix.

[hasan7n]

@sarthakpati The fix will indeed resolve it but I am sure that a sequence of errors will happen after this (I did some hotfixes inside the container but more and more errors are raised). I suggest we understand why it works for you, perhaps I am initializing the repository in an incorrect manner?

One of the things I want to know: so your local machine is picking a brainmage_validation_scores_to_disk.py that allows -md ? or is it picking a CLI_SEGMENT that uses --device for calling brainmage_validation_scores_to_disk?

If the case is the former, could you please point me to the commit hash where brainmage_validation_scores_to_disk allows for -md? I searched the submodules git history multiple times and couldn't find a version like this.

Thanks!

[sarthakpati]

Hmm, that's what I am trying to figure out right now. I'll get back to you.

[sarthakpati]

hey @hasan7n, is this operational and ready to merge?