Create a template bundle with all features tensorboard, mlflow, dice per label

[AHarouni]

Is your feature request related to a problem? Please describe.
Bundles is very powerful and have lots of features but user need to collect pieces of code from different tutorials
It would be nice to have 1 bundle showing all features below

• mlflow --> with flag to disable it if user doesn't need it
• tensorbaord --> with flag to disable it if user doesn't need it
• create different folder for each experiment to same the checkpoint model
• copy json files used into the experiments folder

Describe the solution you'd like

As for the structure we should have

• Have training.json with main training configs including
• Have hypterparamter.json to override hyper parameters
• infer.json to run inference, should also use hyperparamter to get parameters as resolution, intensity range, model dir

this setup should allow user to run multiple experiments just by changing the hyper paramter

Describe alternatives you've considered
I came close to creating this myself but it was lots of trial and error. bundles should provide it all out of the box

[ericspod]

Hi @AHarouni thanks for some great ideas. I think we should do a template bundle that is about bells and whistles like this. For the different folder option, we have this in the template segmentation bundle. For copying the json (or other files) over to this directory, it would be cleaner if more meta information was available to the bundle about itself, that is what arguments were used, what files were loaded, what the current configuration is, etc. I have thought of these enhancements for the bundle code and will discuss these for the 1.5 release.

For the hyperparameter file, I use myself a common.json/yaml file in my bundles defining the network, output directory, base loading transforms, and other things which are common to all "script" configs. I would like this to be standard but this isn't compatible with MONAILabel and other things, so bundles in the zoo haven't done this. We can put this into this proposed bundle anyway and see about future changes to allow it.