read configurations and parameters from file

[msrepo]

Is your proposal related to a problem?

Various configurations and parameters are hardcoded in the code as of now. This means training a new model requires changing these variables scattered around the code. It is also hard to keep track of / document the parameters that were used during a particular instance of training.

Describe the solution you would like.

preferably, a different config.py that consists of multiple python dictionaries. one dictionary each of general(filepaths etc.), network and optimizer-related, and so on.

Additional context

Also, to not have to write too much scaffolding code, may be look at experiment management frameworks.
https://github.com/inferno-pytorch/speedrun-springboard