This project is a description for the pypi package "sklearn-ml", which is for machine learning missions.

pip install sklearn-ml

from sklearn_ml import model_comparison as mc

sklearn-ml supports missions including "binary_classification", "multiclass_classification", "regression" and "clustering", you could choose your mission and models by using the following sample code:

initialized_models = mc.initialize_models(model_names=['knn_c', 'dt_c', 'gbdt_c'], mission='multiclass_classification', seed=2023)

The function returns initialized models that in model_names, you could also customize the model parameters by use parameter "param={model_name:model_parameters}", for example:

params = {'knn_c':{'n_neighbors':15}, 'dt_c':{'max_depth':4}}
initialized_models = mc.initialize_models(model_names=['knn_c', 'dt_c', 'gbdt_c'], mission='multiclass_classification', param=params, seed=2023)

You could train models that are already initiialized in initialized_models by using the following code:

fitted_models, test_results = fit_models(data=X, label=y, models_container=models_container, mission='multiclass_classification')

The function returns the trained models and the test result after using the sklearn function train_test_split() to train and test. Besides, the label y must be categorical, such as [0, 1, 1, 0, ......].

You could also use normalization by setting the "use_normalization" parameter: "use_normalization=minmax" represents MinMaxScaler() for all models, "use_normalization=standard" represents StandardScaler() for all models, a list such as "use_normalization=['minmax', 'standard', 'none', ...]" for models customization is also available.

Here, we use iris dataset from sklearn as an example:

from sklearn.datasets import load_iris
from sklearn_ml import model_comparison as mc 
models_container = mc.initialize_models(model_names=['knn_c', 'dt_c', 'gbdt_c'], mission='multiclass_classification', seed=2023)
# print(models_container)
data = load_iris()
X = data.data
y = data.target
models_fitted, result_df = mc.fit_models(data=X, label=y, models_container=models_container, mission='multiclass_classification')

Once you import sklearn_ml, it will show the usage tips:

We use train_test_split() function to devide training set and testing set. After training the models, we test on the testing set and print the test results