Clara train and python debugging

I’m currently using config files to train/fine_tune my models. Would it be possible to import all these (e.g. pre-transform) classes into Python? So I can easily debug the steps I’m following.
I know there’s ai4med but isn’t this only for when you want to bring-your-own modules? The ai4med is only known inside the Clara train container, i.e. there’s no library .whl library file?

Hi
Thanks for your interest in Clara Train. Unfortunately you can only run train and all transforms from the docker. An easy way to develop is to have interactive session of docker where you mount your code and data. You can then develop with any IDE as pycharm or VScode, Any modifications to the coding folder will be mapped inside the docker and you can run it.
Another alternative is to launch Jupyter lab or notebook (they already exist in the docker) you can use the getting started notebooks as an example of how to do that
https://ngc.nvidia.com/catalog/resources/nvidia:med:clara:getting_started/setup

Hope that helps

Ok, I understand that all Clara Train functionalities are available in the Docker container. I think I wasn’t really clear with my original question: Is it possible to use ai4med library to e.g.: load a Nifti, get a random zoom and save the results. Implementing this example directly into python without using Clara Trains’s “.sh” and “.json” files. Thanks!

Hi
Yes you can use any on the transformations, code in the ai4med libs as long as you adhere to the api inputs and outputs. However, you would be writing/ duplicating lots of code to: load data json/parse, run transforms and save result.
Can you clarify what are you trying to do ? What is missing when using saveasNifti transform