cliyaml 0.1.1

Prototype projects fast with YAML configuration

CliYaml

Prototype projects fast with YAML configuration, override it with CLI.

The idea of this package is to make building reusable, copiable, and configurable scripts in a very fast and readable way.

For instance, this is especially useful when iterating on pytorch model designs and training loops, using separate scripts that can easily be modified, copied, while keeping a readable YAML & CLI api for running them with various configurations.

Basically:

1. Define a typed default config in a subset of YAML (put paths, devices, epochs, learning rates there...)
2. Define cli subcommands that each refer to a config file using the @subcommand("file.yaml") decorator
3. Register subcommands, parse cli args and run the correct subcommands using initialize() and handle()

Usage

YAML subset

This packages relies on YAML for configuration due to its readability, however it only parses a subset of it. Here is what it looks like.

config.yaml

# An example configuration

string: "string"
int: 0
float: 0.0

# docstring
docstring: "value"

bool: true

empty: # int

nested:
  one: 1
  two: 2

• the top comments are the subcommand's description
• comments before a value are its docstring
• only true and false are valid booleans
• type hints can be specified for null values with an inline comment
• no lists
• nested values are flattened into a flat python dict, with keys joined by _

CLI API

Define some scripts in separate files, for instance in a scripts/ folder.

scripts/main.py:

from cliyaml import subcommand

# NOTE: multiple different files can be specified to extend the configuration
@subcommand("config.yaml")
def main(**kwargs):
    print("Called with args:", kwargs)

In your main.py, include the following code:

if __name__ == "__main__":
    import cliyaml

    # Registers subcommands
    # NOTE: you can also register single files
    cliyaml.initialize(None, "scripts/")

    # Parses CLI args and runs commands
    cliyaml.handle()

Then, run the following code to use your API:

python main.py -h
python main.py main -h
python main.py main

You can also use cliyaml.call() to call a function by giving it kwargs directly, it will only pass the correct arguments. This function also allows you to pass named and list arguments to override the kwargs.

Building

uv build

Publish

uv publish

Test

pytest

TODO : more comprehensive tests