Usage

An example of a JSON configuration file could look like this [1] :

{
  "architecture": {
    "convNet": {
      "spectrograms": {
        "frame_sizes": [
          1024,
          2048,
          4096
        ],
        "hop_size": 441,
        "num_bands": 40
      },
      "training_data": {
        "test_size": 0.1,
        "val_size": 0.2,
        "random_state": 42
      },
      "paths": {
        "files": {
          "train_logs": "train_logs.csv",
          "state": "state.json",
          "model": "model.hdf5"
        },
        "dirs": {
          "store": "/path/to/data_store",
          "sandbox": "/path/to/sandbox",
          "tests": "/path/to/test_dir"
        }
      },
      "overwrite": false,
      "default_params": null
    }
  }
}

Note that these configurations can be nested. You can use strings, floats, integers, lists, booleans and null values. The JSON format does not support None directly but, once imported, null will be converted to None. The same goes for false/true which will be converted to their upper case equivalents False and True.

To use dot-configs in your project:

import dot_configs as dc
config_filepath = '/path/to/json_configuration_file'
cfg = dc.Configurations(config_filepath).get_configurations()

Now you can address any of the values in the following way:

cfg.architecture.convNet.paths.files.train_logs    # "train_logs.csv"

Footnotes

[1]	You can find this example JSON under examples in the repository.