easy-ml-config 1.1.0

provides a convenient base configuration class for machine learning projects

easy-ml-config

Super lightweight library to provide a simple base configuration class to maintain consistency and reproducibility for machine learning projects. There is only one module ml_config.configuration that contains the BaseConfig class, which allows you to:

• Define complex and nested configuration classes.
• Export/Import your configuration to a dictionary or a yaml file with simply to_dict(), from_dict, to_yaml and from_yaml.

from dataclasses import dataclass
from easy_ml_config import BaseConfig

@dataclass
class ValidationConfig(BaseConfig):
    batch_size: int
    upload_to_server: bool = False

@dataclass
class TrainConfig(BaseConfig):
    num_epochs: int
    batch_size: int
    learning_rate: float
    validation_config: ValidationConfig | None = None

class ModelConfig(BaseConfig):
    num_layer: int

@dataclass
class MyExpConfig(BaseConfig):
    model: MyModelConfig
    train: TrainConfig
    exp_id: str

# Initialize from a dictionary
exp_config = MyExpConfig.from_dict({
    "model": {
        "num_layer": 3,
    },
    "train": {
        "num_epochs": 10,
        "batch_size": 32,
        "learning_rate": 0.001,
        "validation_config": {
            "batch_size": 16,
            "upload_to_server": False
        } # optional
    },
    "exp_id": "exp_1"
})

# Export to a yaml file
exp_config.to_yaml("exp_config_001.yaml")

# Initialize from a yaml file
exp_config = MyExpConfig.from_yaml("exp_config_001.yaml")

The yaml file will look like this:

model:
    num_layer: 3
train:
    num_epochs: 10
    batch_size: 32
    learning_rate: 0.001
    validation_config:
        batch_size: 16
        upload_to_server: false
exp_id: exp_1

Installation

pip install easy-ml-config

Usage

Start with defining your own configuration by inheriting from BaseConfig. Once you have that, you can utilize all the predefined functionalities of BaseConfig.

Define Your Configuration

from dataclasses import dataclass

from easy_ml_config import BaseConfig


@dataclass
class MyModelConfig(BaseConfig):
    num_layer: int

@dataclass
class MyExpConfig(BaseConfig):
    model: MyModelConfig
    batch_size: int
    learning_rate: float

Instantiate Your Configuration

When you are running an experiment with a specific settings. Initialize your configurations.

Initialize from A Dictionary

def main():
    nested_args = {
        "model": {
            "num_layer": 3,
        },
        "batch_size": 32,
        "learning_rate": 0.001,
    }
    config = MyExpConfig.from_dict(nested_args)
    # model = MyModel(num_layer=config.model.num_layer)

if __name__ == "__main__":
    main()

Initialize from A YAML File

# config.yaml
model:
  num_layer: 3
batch_size: 32
learning_rate: 0.001

config = MyExpConfig.from_yaml("config.yaml")

Output Your Configuration

Often you might want to save your configuration to a file. You can do that easily with the to_dict and to_yaml methods.

# Save to a dictionary
config_dict = config.to_dict()
# Save to a yaml file
config.to_yaml("config.yaml")

The exported dictionary and yaml file can be directly used again to recreate the Configurations.

# Load from a dictionary
config = MyExpConfig.from_dict(config_dict)
# Load from a yaml file
config = MyExpConfig.from_yaml("config.yaml")