# Dataconfigs
## About
Dataconfigs provides a seamless integration between dataclasses as configs and any configurable classes that use these configs. The most powerful feature is that config attributes become part of configurable, i.e., no need for _explicit_ `.config` attribute (unless specified otherwise). This keep class definitions very siple yet decouples configurable parameters from class attributes. In terms of configs, it is a simplified hydra alternative, yet supporting more features.
## Features
Last 3 features are in development:
* Embedded config params in configurables (no need for separate `.config` attribute, unless specified otherwise)
* Easy type annotations (visible annotations, even though configs are never inherited from)
* Support for multiple, nested, hierarchical, and union configs, as well as non-atomic types
* Various file support (`json`, `yaml`, `pickle`)
* CLI support
* Decorator-based (inspired by Hydra)
## Quick Start
Install the library:
```bash
pip install dataconfigs
```
Check the [demo](https://github.com/mantasu/dataconfigs/blob/main/notebooks/01-demo.ipynb).
### Simple Script Example
1. Simply define your config (must be a datacalss and contain the word Config in name!)
2. Extend the config with your configurable (it won't turn into a dataclass unless specified!)
3. Instantiate your configurable by optionally overwriting default config parameters
```python
from datacalsses import dataclass, is_dataclass
from dataconfigs import configurable, show_config_params
@datacalss
class MyConfig:
"""My config class
Args:
param1 (int): The first parameter.
param2 (str): The second parameter.
"""
param1: int = 0
param2: str = "param2"
@configurable
class MyConfigurable(MyConfig):
def __init__(self, attr1=1, attr2=2):
print(self.param1, self.param2) # Already available!
assert not is_dataclass(MyConfigurable) # Only attributes were copied!
obj = MyConfigurable("attr1", param1=1) # Can overwrite config params!
# 1, 'param2'
```
### Simple CLI Example (only ideas for now)
Assume you have `config.json` file as follows:
```json
{
"param1": 111,
"param2": "222"
}
```
One possibility is to directly make the configurable. Example `main.py`:
```python
@configurable.main("path/to/config.json", MyConfigurable, attr1="attr1")
def main(my_configurable):
pass
if __name__ == "__main__":
main()
```
We can still overwrite config params or the whole config file when calling:
```bash
dataconfigs main.py --param2 'custom' # 111, 'custom'
dataconfigs main.py --config path/to/config2.yaml
```
If we call `dataconfigs main.py -h`, descriptions will be extracted from docstrings:
```bash
Options:
--param1 : (int) The first parameter. Defaults to 111.
--param2 : (str) The second parameter. Defaults to '222'.
```
## Tutorials (TODO)
1. **Config classes**: there are many ways to define configs. The only restriction is that they must be dataclasses and contain an "Config" word in their class name (see basics):
* **Basics**
* **Unions & non-atomic types**
* **Multiple, nested and hierarchical configs**
* **Config files (json, yaml, pkl, etc.)**
2. **Configurables**: any class (configurable) that extends the config(-s) will only obtain its attributes and won't alter its class definition. The configurable class can inherit from any other parents in the usual manner.
* **Config vs non-config attributes**
* **Various ways to instantiate**
3. Compatibility with Third Party CLI
* Pytorch Lightning
* Tensorflow
* Optuna
* Ray
Raw data
{
"_id": null,
"home_page": "https://github.com/mantasu/dataconfigs",
"name": "dataconfigs",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.12",
"maintainer_email": null,
"keywords": "python, config, configuration, configurable",
"author": "Mantas Bir\u0161kus",
"author_email": "mantix7@gmail.com",
"download_url": "https://files.pythonhosted.org/packages/03/1c/a3a550e314a7f0ee90dd42535643dc6adafd60f9efd11724cc460ce88881/dataconfigs-0.1.0.tar.gz",
"platform": null,
"description": "# Dataconfigs\n\n## About\n\nDataconfigs provides a seamless integration between dataclasses as configs and any configurable classes that use these configs. The most powerful feature is that config attributes become part of configurable, i.e., no need for _explicit_ `.config` attribute (unless specified otherwise). This keep class definitions very siple yet decouples configurable parameters from class attributes. In terms of configs, it is a simplified hydra alternative, yet supporting more features.\n\n## Features\n\nLast 3 features are in development:\n* Embedded config params in configurables (no need for separate `.config` attribute, unless specified otherwise)\n* Easy type annotations (visible annotations, even though configs are never inherited from)\n* Support for multiple, nested, hierarchical, and union configs, as well as non-atomic types\n* Various file support (`json`, `yaml`, `pickle`)\n* CLI support\n* Decorator-based (inspired by Hydra)\n\n## Quick Start\n\nInstall the library:\n\n```bash\npip install dataconfigs\n```\n\nCheck the [demo](https://github.com/mantasu/dataconfigs/blob/main/notebooks/01-demo.ipynb).\n\n### Simple Script Example\n\n1. Simply define your config (must be a datacalss and contain the word Config in name!)\n2. Extend the config with your configurable (it won't turn into a dataclass unless specified!)\n3. Instantiate your configurable by optionally overwriting default config parameters\n\n```python\nfrom datacalsses import dataclass, is_dataclass\nfrom dataconfigs import configurable, show_config_params\n\n@datacalss\nclass MyConfig:\n \"\"\"My config class\n Args:\n param1 (int): The first parameter.\n param2 (str): The second parameter.\n \"\"\"\n param1: int = 0\n param2: str = \"param2\"\n\n@configurable\nclass MyConfigurable(MyConfig):\n def __init__(self, attr1=1, attr2=2):\n print(self.param1, self.param2) # Already available!\n\nassert not is_dataclass(MyConfigurable) # Only attributes were copied!\nobj = MyConfigurable(\"attr1\", param1=1) # Can overwrite config params! \n# 1, 'param2'\n```\n\n### Simple CLI Example (only ideas for now)\n\nAssume you have `config.json` file as follows:\n```json\n{\n \"param1\": 111,\n \"param2\": \"222\"\n}\n```\n\nOne possibility is to directly make the configurable. Example `main.py`:\n```python\n@configurable.main(\"path/to/config.json\", MyConfigurable, attr1=\"attr1\")\ndef main(my_configurable):\n pass\n\nif __name__ == \"__main__\":\n main()\n```\n\nWe can still overwrite config params or the whole config file when calling:\n```bash\ndataconfigs main.py --param2 'custom' # 111, 'custom'\ndataconfigs main.py --config path/to/config2.yaml\n```\n\nIf we call `dataconfigs main.py -h`, descriptions will be extracted from docstrings:\n```bash\nOptions:\n--param1 : (int) The first parameter. Defaults to 111.\n--param2 : (str) The second parameter. Defaults to '222'.\n```\n\n## Tutorials (TODO)\n\n1. **Config classes**: there are many ways to define configs. The only restriction is that they must be dataclasses and contain an \"Config\" word in their class name (see basics):\n * **Basics**\n * **Unions & non-atomic types**\n * **Multiple, nested and hierarchical configs**\n * **Config files (json, yaml, pkl, etc.)**\n2. **Configurables**: any class (configurable) that extends the config(-s) will only obtain its attributes and won't alter its class definition. The configurable class can inherit from any other parents in the usual manner.\n * **Config vs non-config attributes**\n * **Various ways to instantiate**\n3. Compatibility with Third Party CLI\n * Pytorch Lightning\n * Tensorflow\n * Optuna\n * Ray\n\n",
"bugtrack_url": null,
"license": "Apache",
"summary": "Turn your dataclasses into configs seamlessly.",
"version": "0.1.0",
"project_urls": {
"Bug Tracker": "https://github.com/mantasu/dataconfigs/issues",
"Documentation": "https://mantasu.github.io/dataconfigs",
"Homepage": "https://github.com/mantasu/dataconfigs"
},
"split_keywords": [
"python",
" config",
" configuration",
" configurable"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "51ba83c5e928c50d97cbaf5ab50ccac889e2d2d9d0ceae7fde0175dbd992f48c",
"md5": "e4eae7fc433148f3d158a617bfd7e0f6",
"sha256": "ac494a18b6e2b31aae23f60c56676f9b218007883386eaf74cc7352e65b53331"
},
"downloads": -1,
"filename": "dataconfigs-0.1.0-py3-none-any.whl",
"has_sig": false,
"md5_digest": "e4eae7fc433148f3d158a617bfd7e0f6",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.12",
"size": 18360,
"upload_time": "2024-10-28T02:44:51",
"upload_time_iso_8601": "2024-10-28T02:44:51.528813Z",
"url": "https://files.pythonhosted.org/packages/51/ba/83c5e928c50d97cbaf5ab50ccac889e2d2d9d0ceae7fde0175dbd992f48c/dataconfigs-0.1.0-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "031ca3a550e314a7f0ee90dd42535643dc6adafd60f9efd11724cc460ce88881",
"md5": "99909df91a0cf9d730dfc2dfffe9d55c",
"sha256": "3c56f0be32762dd1be33dec439e883b87e2d5e8088f4e23a910b1a0f5c545355"
},
"downloads": -1,
"filename": "dataconfigs-0.1.0.tar.gz",
"has_sig": false,
"md5_digest": "99909df91a0cf9d730dfc2dfffe9d55c",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.12",
"size": 18347,
"upload_time": "2024-10-28T02:44:52",
"upload_time_iso_8601": "2024-10-28T02:44:52.721559Z",
"url": "https://files.pythonhosted.org/packages/03/1c/a3a550e314a7f0ee90dd42535643dc6adafd60f9efd11724cc460ce88881/dataconfigs-0.1.0.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-10-28 02:44:52",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "mantasu",
"github_project": "dataconfigs",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"requirements": [],
"lcname": "dataconfigs"
}
JSON
