configgle 0.1.1

Tools for making configurable Python classes for A/B experiements.

configgle 🤭

Tools for making configurable Python classes for A/B experiements.

Installation

python -m pip install configgle

Example

from configgle import Fig

class Model:
    class Config(Fig):
        hidden_size: int = 256
        num_layers: int = 4

    def __init__(self, config: Config):
        self.config = config

# Create and modify config
config = Model.Config(hidden_size=512)

# Instantiate the parent class
model = config.setup()
print(model.config.hidden_size)  # 512

Or use @autofig to auto-generate the Config from __init__:

from configgle import autofig

@autofig
class Model:
    def __init__(self, hidden_size: int = 256, num_layers: int = 4):
        self.hidden_size = hidden_size
        self.num_layers = num_layers

# Config is auto-generated from __init__ signature
model = Model.Config(hidden_size=512).setup()
print(model.hidden_size)  # 512

References

configgle combines Python standard patterns (dataclasses) with hierarchical, class-local configuration.

The following libraries span these ideas but none wholly combine them:

• Confugue - Hierarchical configuration with YAML-based object instantiation (most similar to configgle, but uses YAML rather than pure Python)
• Fiddle - Python-first configuration library for ML
• Gin Config - Lightweight configuration framework for Python
• Sacred - Tool to configure, organize, log and reproduce experiments
• Hydra - Framework for elegantly configuring complex applications
• ml_collections - Python collections designed for ML use cases
• OmegaConf - Flexible hierarchical configuration system

License

Apache License 2.0