Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/hmasdev/sklearn-protocols
sklearn-protocols: protocols for sklearn regressors, classifiers and transformers
https://github.com/hmasdev/sklearn-protocols
protocol python python3 scikit-learn
Last synced: about 22 hours ago
JSON representation
sklearn-protocols: protocols for sklearn regressors, classifiers and transformers
- Host: GitHub
- URL: https://github.com/hmasdev/sklearn-protocols
- Owner: hmasdev
- License: mit
- Created: 2022-01-30T14:56:22.000Z (almost 3 years ago)
- Default Branch: master
- Last Pushed: 2022-01-30T15:01:42.000Z (almost 3 years ago)
- Last Synced: 2023-08-25T19:42:00.973Z (about 1 year ago)
- Topics: protocol, python, python3, scikit-learn
- Language: Python
- Homepage:
- Size: 8.79 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Sklearn-Protocols: protocols for sklearn regressors, classifiers and transformers
Scikit-learn is one of the most commonly used machine-learning tools in python.
However, scikit-learn has a difficulty in type hints.
For example, when you write
```python
import numpy as np
from sklearn.base import BaseEstimator
def train(model: BaseEstimator, X: np.ndarray, y:np.ndarray) -> BaseEstimator:
model.fit()
return model
def train_(model: RegressorMixin, X: np.ndarray, y:np.ndarray) -> BaseEstimator:
model.fit()
return model
```
in order to train various kinds of models in a regression task, `model.fit` is not appropriately inferred with python type-checking system in Visual Studio Code.
This is because `fit` method is not included in `BaseEstimator` nor `RegressorMixin`.
It is also a reason that there are not protocols or interfaces in scikit-learn.
So, in order to solve this problem, `sklearn_protocols` provides sklearn-compatible protocols for regressors, classifiers, transformers and so on.
NOTE: [This thread](https://stackoverflow.com/a/60542986/165678329) will help you understand `sklearn_protocols`.
## Installation
### Requirements
- python >= 3.8
- scikit-learn
### User Instllation
```bash
pip install git+https://github.com/hmasdev/sklearn-protocols.git@master
```
### Basic Protocols
`sklearn_protocols` has five protocols as the basic protocols:
1. `RegressorProtocol`;
2. `ClassifierProtocol`;
3. `ClassifierWithPredictProbaProtocol`;
4. `TransformerProtocol`;
5. `ClusterProtocol`.
The definitions of those protocols are follows:
1. `RegressorProtocol` is the interface which prediction models should satisfy in regression tasks. This protocol has `get_params`, `set_params`, `fit`, `predict` and `score` as its methods;
2. `ClassifierProtocol` is the interface which prediction models should satisfy in classification tasks. This protocol has `get_params`, `set_params`, `fit`, `predict` and `score` as its methods;
3. `ClassifierWithPredictProbaProtocol` is the interface which prediction models should satisfy in classification tasks. However, `ClassifierWithPredictProbaProtocol` is different from `ClassifierProtocol`. The former has `predict_proba` method but the latter does not. That is, this protocol has `get_params`, `set_params`, `fit`, `predict` `score` and `predict_proba` as its methods;
4. `TransformerProtocol` is the interface which preprocessors should satisfy. This protocol has `get_params`, `set_params`, `fit`, `transform` and `fit_transform`;
5. `ClusterProtocol` is the interface which clustering models should satisfy in clustering tasks. This protocol has `get_params`, `set_params`, `fit` and `fit_predict`.
### How to Use
`sklearn_protocols` provides only protocols.
So, you can use the protocols as follows.
1. Type Hint:
For example, in regression tasks, you can use RegressorProtocol like
```python
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn_protocols import RegressorProtocol
def train(
model: RegressorProtocol,
X: np.ndarray,
y: np.ndarray,
) -> RegressorProtocol:
model.fit(Xtr, ytr) # RegressorProtocol has `fit`.
return model
```
2. Type Check:
You can use `isinstance` or `issubclass` with the protocols to find out whether an object has methods required in machine learning tasks.
Also, the protocols are mypy-compatible.
```python
my_model = get_model()
if isinstance(my_model, RegressorProtocol):
print("my_models is a regressor")
```
If the protocols, `RegressorProtocol`, `ClassifierProtocol`, `ClassifierWithPredictProbaProtocol`, `TransformerProtocol` and `ClusterProtocol`, are insufficient for your requirements, you can create a custom protocol with `sklearn_protocols.base` and `sklearn_protocols.support`:
- `sklearn_protocols.base`;
- `BaseEstimatorProtocol`: compatible with `sklearn.base.BaseEstimator`;
- `BiclusterMixinProtocol`: compatible with `sklearn.base.BiclusterMixin`;
- `ClassifierMixinProtocol`: compatible with `sklearn.base.ClassifierMixin`;
- `ClusterMixinProtocol`: compatible with `sklearn.base.ClusterMixin`;
- `DensityMixinProtocol`: compatible with `sklearn.base.DensityMixin`;
- `RegressorMixinProtocol`: compatible with `sklearn.base.RegressorMixin`;
- `TransformerMixinProtocol`: compatible with `sklearn.base.TransformerMixin`;
- `SelectorMixinProtocol`: compatible with `sklearn.feature_selection.SelectorMixin`;
- `sklearn_protocols.support`;
- `SupportFit`: : only supporting `fit` method;
- `SupportPredict`: only supporting `predict` method;
- `SupportPredictProba`: only supporting `predict_proba` method;
- `SupportPredictLogProba`: only supporting `predict_log_proba` method;
- `SupportTransform`: only supporting `transform` method;
- `SupportFitPredict`: only supporting `fit_predict` method;
- `SupportFitTransform`: only supporting `fit_transform` method.
## Contribution Guide
### Requirements
- python >= 3.8
- pipenv
### Setup
After `fork`,
```bash
$ git clone https://github.com/{YOUR_ACCOUNT}/sklearn-protocols.git
$ cd sklearn-protocols
$ pipenv install --dev
```
### Issues
- For any bugs, use [BUG REPORT](https://github.com/hmasdev/sklearn-protocols/issues/new?assignees=&labels=bug&template=bug_report.md&title=%5BBUG%5D) to create an issue.
- For any enhancement, use [FEATURE REQUEST](https://github.com/hmasdev/sklearn-protocols/issues/new?assignees=&labels=enhancement&template=feature_request.md&title=) to create an issue.
- For other topics, create an issue with a clear and concise description.
### Pull Request
1. Fork (https://github.com/hmasdev/sklearn-protocols/fork);
2. Create your feature branch (git checkout -b feautre/xxxx);
3. Test codes according to Test Subsection;
4. Commit your changes (git commit -am 'Add xxxx feature);
5. Push to the branch (git push origin feature/xxxx);
6. Create new Pull Request
### Test
```bash
$ pipenv run flake8
$ pipenv run mypy .
$ pipenv run pytest
```
## LICENSE
MIT
## Authors
[hmasdev](https://github.com/hmasdev)