ClearConf is a library created to support easy and manageble python configuration. It consists in a python API (based on the BaseConfig class) which adds additional functionalities to a configuration class, and in an optional CLI tool to simplify managing configurations.

To install ClearConf just run

pip install clearconf

A configuration file for a machine learning project could be structure as follow.

from models import MLP
from datasets import ImageNet


class Config(BaseConfig):
    seed = 1234

    class Model:
        architecture = MLP

        class Params:
            num_layers = 16
            layers_dim = [96] * num_layers


    class Data:
        dataset = ImageNet

        class Params:
            root = './data/PCN'
            split = 'PCN.json'
            subset = 'train'

The training/test script can access the configuration simply by importing it:

from configs import Config

Model = Config.Model
Data = Config.Data

model = Model.architecture(**Model.Params.to_dict())
dataset = Data.dataset(**Data.Params.to_dict())

It is also possible to simplify the configuration further using inheritance. For example the Model configuration seen before would look like this:

from models import MLP

class Config(BaseConfig):
    seed = 1234

    class Model(MLP):
        num_layers = 16
        layers_dim = [96] * num_layers

The corresponding script is:

from configs import Config

model = Config.Model()

The first step to use clearconf would be to use the CLI tool in the root of your project to initialize it:

cconf init

This will generate a config directory where you will store your configurations and a .clearconf file used by ClearConf to keep track of configurations. After this you can start populating your config directory. You can find examples of configuration files in the Example section.

Finally you can import a generic configuration in your script as

from configs import Config

and use it as you please.

When the script is run, if a default configuration has been set via the CLI

cconf defaults add main.py test_conf.py

such configuration will be dynamically imported.

Otherwise, clearconf will list all the available configuration and ask you to pick one.

0: example3_conf
1: example1_conf
2: example2_conf
3: example4_conf
Choose a configuration file: