hackerargs 1.1.1

Easy config system, minimizing new lines of code

hackerargs

hackerargs minimizes new lines of code to add a configurable argument anywhere in your codebase, so you can focus on coding.

Hackerargs is for everyone, but is especially designed for researchers using config-heavy ML libraries like pytorch, coding largely from scratch in solo or small-team repos, and running experiments varying CLI args with wandb.

# in any python file in your codebase
from hackerargs import args

class Model:
    def __init__(self):
        self.parameter = args.setdefault(key, default_value)
        # run forth and code with confidence!
        # value is taken from CLI or yaml config if given

• In hackerargs, args is a global, write-once-only dict
• We emphasize val = args.setdefault(key, default_val) as a primitive. It returns your key's value if it already exists, and also sets it to default_val if missing. No more fumbling with errors accessing missing keys. Write-once only means initialized args from CLI or yaml config are used at runtime, and exact reproducibility on reruns by reloading saved yaml config
• Write cleaner code, simplify function parameters, and operate at a higher level of abstraction: no more passing args around, or long function signatures filled with low-level details

Features

• Type inference as floats, ints, strings, lists, etc. with PyYAML loader (YAML v1.1), except "on/off/yes/no" are not parsed as booleans
• Optional integration with argparse
• Works with wandb sweeps: log config using wandb.config.update(args), and run sweep experiments in CLI --{key} {val} format

Installation

Hackerargs is a pip package and conda package:

pip install hackerargs

or

conda install mxwsn::hackerargs
conda install -c mxwsn hackerargs

Initialization

Initialize hackerargs in your driver script, just like argparse:

from hackerargs import args

if __name__ == '__main__':
    args.parse_args()
    main()
    args.save_to_yaml(yaml_file)

parse_args can be called with no arguments:

• args.parse_args(): Parse CLI options in --{key} {val} format. Then, if --config {yaml_file} provided, load from yaml file.

parse_args can also take either a YAML config file, or argparse.ArgumentParser.

• args.parse_args('config.yaml')
• args.parse_args(argparse.ArgumentParser())
• args.parse_args(argparse.ArgumentParser(), 'config.yaml')
• args.parse_args('config.yaml', argparse.ArgumentParser())

parse_args parses sys.argv by default, but you can use a custom argv instead:

• args.parse_args(..., argv = ['--string', 'text', '--int', '42'])

Priority

In general, CLI options > yaml config (if provided).

1. (Highest priority) ArgumentParser options specified by user via CLI, if ArgumentParser is given
2. Unknown CLI options (not recognized by ArgumentParser) specified by user. These are parsed in --{key} {val} format. If no ArgumentParser is given, then all CLI options are parsed this way
3. YAML config. If --config {yaml_file} CLI option is given, it is used instead of a yaml file given as input to parse_args() in python. As such, --config is a protected CLI option when using hackerargs
4. (Lowest priority) ArgumentParser default values for options not specified by user via CLI, if ArgumentParser is given

As a write-once-only dict, initialized values take priority over runtime values.

Access

Running python train.py --string text --int 42 --float 3.14 --list [1,2], we have:

>>> args
{'string': 'text', 'int': 42, 'float': 3.14, 'list': [1, 2]}
>>> args.setdefault('float', 1e-4)
3.14
>>> args.get('float')
3.14
>>> args['float']
3.14
>>> args.float
3.14

When a key might not be set, we recommend using setdefault for access. When you're sure the key is set already, hackerargs supports .get, bracket access, and dot access.

Compatibility with argparse

Build your own ArgumentParser, and pass it to hackerarg:

import argparse
from hackerargs import args

if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--example', description = 'Example argument')
    parser.add_argument('--flag', action = 'store_true')
    # ...

    args.parse_args(parser)               # Uses parser on CLI args

hackerargs will first use your ArgumentParser to parse known arguments, then parse unknown CLI arguments in --{key} {val} format.

Running python example.py --example text --unknown 1, we get:

args = {'example': 'text', 'flag': False, 'unknown': 1}

Supporting argparse lets you create help messages for your scripts easily.