Clarifying requirements for training from local hierarchy

[DrRajJena]

Folks. I followed the instructions for creating a local hierarchy to create new training runs. I copied what I thought was the minimum number of files needed.

My tree listing for subfolder InnerEyeFoo in InnerEyeDeepLearning
├── ML
│ ├── configs
│ │ └── segmentation
│ │ ├── HeadAndNeckBase.py
│ │ ├── ProstateAB.py
│ │ └── ProstateBase.py
│ └── runner.py
└── settings.yml

My settings.yml looks like this
subscription_id:
application_id: ''
azureml_datastore:
resource_group:
docker_shm_size: '440g'
workspace_name:
cluster: 'NC24'
model_configs_namespace: 'InnerEyeFoo.ML.configs'
extra_code_directory: 'InnerEyeFoo'

Running in the InnerEye conda environment, my runner.py cannot see InnerEye libraries, even though I can manually import them from the command line from the same path.

Could you advise if I have missed steps to create my local hierarchy?

[ant0nsc]

Hi @DrRajJena, I can't say for certain how to resolve that without knowing where your InnerEye folder lives. Is your setup like

InnerEye-DeepLearning
   > InnerEye
   > Tests
   > InnerEyeFoo
         > ML
             runner.py

If so, you would need to modify sys.path inside of runner.py before you make your first import statement from the InnerEye codebase. It would look something like:

import sys
from pathlib import Path
# Here, you need to adjust how often you call .parent to get to the root of the project where the InnerEye folder is located.
root = Path(__file__).resolve().parent.parent

if (root / "InnerEye").is_dir():
    root_str = str(root)
    if root_str not in sys.path:
        print(f"Adding to sys.path: {root_str}")
        sys.path.insert(0, root_str)
else:
    raise ValueError("I'm expecting to find a folder called InnerEye in the parent folder of this file.")

[DrRajJena]

Perfect - thank you!