hiargparse 0.2.2

A hierarchical and highly sophisticated variant of argparse.

Hierarchical Argparse

Hiargparse is a hierarchical and highly sophisticated variant of argparse.

Minimal Code

from hiargparse import ArgsProvider, Arg, ChildProvider, ArgumentParser

child = ArgsProvider(
    args=[Arg('baz', default=42)]
)
root = ArgsProvider(
    args=[Arg('foo', default='bar')],
    child_providers=[ChildProvider(provider=child, name='child')]
)
parser = ArgumentParser()
parser.add_arguments_from_provider(root)
print(parser.parse_args())

Description

Hiargparse automatically generates the command-line arguments with your all classes in your tree-like module structures with minimal codings.

Suppose you make a large module, and your hierarchically deep class, say foo.bar.baz.ham.spam.Egg, requires some arguments, say heights and widths. Foo makes an instance of Bar, Bar makes an instance of Baz, ..., and Spam makes an Egg which has the two arguments. When you want to pass them command-line options, you have to write the two, heights and widths, in the constructor of Foo, Bar, ..., and Spam. What if the Egg is updated and wants to require some new arguments, like depth or length? Hiargparse passes the arguments directly to the classes without any black magics.

Features

With this module, you can

• easily make hierarchical (tree-like) command-line arguments with argparse
• Each argument is automatically help-texted and grouped into argparse.ArgumentGroup.
• get a more useful Namespace object than the original
• Accessing with dict-like key, getting the child Namespace, converting to/from dictionaries, and so on
• write/read the arguments to/from some configure files with well known formats
• Currently we supports yaml and toml.

Also, this module

• is almost compatible with original argparse; you can gradually introduce it to your large projects.
• works without command-line arguments; when you distribute your module with hiargparse, users still can select whether to feed arguments to it with a command-line or programatic way.

This module is inspired by rlpytorch.args_provider in FacebookAIResearch/ELF.

Installation

pip install hiargparse

Requirements

• python >= 3.6.0
• typing_extensions (for typing_extentions.Protocol in python 3.6) >= 3.6.2.1
• pyyaml >= 3.12 (only if you use yaml write/read)
• toml >= 0.9.4 (only if you use toml write/read)

Examples

• See /examples/example.py to easy start.
• You can use hiargparse with original ArgumentParser. See /examples/example_with_original.py.
• If you want to use hiargparse without command-line arguments, see /examples/example_without_program_arguments.py.
• /examples/example_write_and_read.py describes how to write and read the arguments with a configure file.
• A lot of things you can do with hiargparse are shown in /examples/complicated_example.py.

Contribution

• Any contribution is welcome!
• Fork the repo, create a branch, add your awesome changes, and make a new Pull Request.
• If you find some bugs, please report in issue.

Author

• Keigo Kawamura (Department of Electrical Engineering and Information Systems (EEIS), Graduate School of Engineering, The University of Tokyo)
• [e-mail]