Argparse code can be longer than actual implementation code!
That's a problem I find with most popular argument parsing options is that if your parameters are only modest, the code to document them becomes disproportionately large to the benefit they provide.
A relative new-comer to the argument parsing scene (I think) is plac.
It makes some acknowledged trade-offs with argparse, but uses inline documentation and wraps simply around main()
type function function:
def main(excel_file_path: "Path to input training file.",
excel_sheet_name:"Name of the excel sheet containing training data including columns 'Label' and 'Description'.",
existing_model_path: "Path to an existing model to refine."=None,
batch_size_start: "The smallest size of any minibatch."=10.,
batch_size_stop: "The largest size of any minibatch."=250.,
batch_size_step: "The step for increase in minibatch size."=1.002,
batch_test_steps: "Flag. If True, show minibatch steps."=False):
"Train a Spacy (http://spacy.io/) text classification model with gold document and label data until the model nears convergence (LOSS < 0.5)."
pass # Implementation code goes here!
if __name__ == '__main__':
import plac; plac.call(main)