yacrs 0.1.0

Yet Another Configuration and Registration System

YACRS: Yet Another Configuration and Registration System

YACRS is a lightweight Python configuration system that maps configuration values directly to class or function arguments. It is inspired by YACS but adds a decorator-based registration and binding mechanism so you can wire configs to existing code without writing boilerplate mapping logic.

Philosophy

Get rid of the configuration value mapping in code blocks.

Features

• Hierarchical config nodes — Node extends dict and supports dotted-key access (e.g. cfg.model.lr).
• Auto-vivification — Nested paths are created on demand.
• Immutable snapshots — Freeze a config tree to prevent accidental changes.
• Decorator-based binding — @register maps config values to function/class __init__ arguments.
• Scope binding — Bind the same class/function to different config scopes.
• CLI support — Load JSON/YAML/TOML configs and override values from the command line.

Installation

pip install yacrs

Or install from source:

git clone https://github.com/Cowhisper/yacrs.git
cd yacrs
pip install -e .

Quick Start

from yacrs import _C, register

# 1. Register a class with a config scope
@register('model')
class Model:
    def __init__(self, input_channels, output_channels):
        self.input_channels = input_channels
        self.output_channels = output_channels

# 2. Create config entries
_C.register('model')
_C.model.input_channels = 3
_C.model.output_channels = 32

# 3. Instantiate from config
model = Model()
print(model.input_channels)  # 3
print(model.output_channels)  # 32

See tutorial.md for a complete walkthrough.

Core Concepts

Node

A Node is a dictionary that supports attribute-style and dotted-key access:

from yacrs import Node

cfg = Node()
cfg.model = Node()
cfg.model.lr = 0.01
cfg.set('model.optimizer', 'Adam')

print(cfg.get('model.lr'))  # 0.01
print(cfg.model.optimizer)  # Adam

_C — Global Config

_C is the package-wide root Node. It is where @configurable looks up values by default.

@configurable

The configurable decorator registers a class or function and binds its parameters to config values at call time.

from yacrs import _C, register

@register('train')
def train(epoch, lr, model):
    print(f'Training {model} for {epoch} epochs at lr={lr}')

_C.register('train')
_C.train.epoch = 10
_C.train.lr = 0.001
_C.train.model = 'resnet50'

train()  # Training resnet50 for 10 epochs at lr=0.001

CLI

Use cli(...) to load config files and override values from the command line:

python train.py -c config.yaml train.epoch=20 train.lr=0.01

See tutorial.md for details on CLI usage and config file formats.

License

MIT