theOneMLFunction

This Python script provides a one-stop machine learning function that can handle both regression and classification tasks. It includes data cleaning, feature selection, model training, hyperparameter tuning, result printing, model ensembling.

This Python module includes theOneMLFunction, a machine learning function designed for a variety of classification and regression tasks. The function accepts a data set, target variable, and a binary indicator of whether the target variable is continuous or not. The function cleans the data, encodes categorical features, selects the most important features, creates a stratified subset, trains a set of candidate models and tunes their hyper-parameters. It also ensembles different combinations of the top three best models. Finally, the function retrains the best model on the whole data and returns the best model and the list of selected features.

Installation

This function requires Python 3.7 or later and the following dependencies:

scikit-learn
numpy
pandas
scipy

To install the required packages, you can use pip:

pip install scikit-learn numpy pandas scipy

Usage:

from theOneMLFunction import theOneMLFunction

# data: pandas.DataFrame containing input data
# targetVar: str containing the name of the target variable in data
# isContinuous: bool indicating whether the problem is a regression problem (True) or a classification problem (False)
# randomState: int setting the seed for the random number generator for reproducibility purposes

bestModel, selectedFeatures = theOneMLFunction(data, targetVar, isContinuous, randomState)

The function returns the best trained model and the selected features, which are the 5 most important features selected using a feature selection method.

Details

The function currently uses mutual information-based feature selection, and it is not currently customizable.
The following candidate models are currently included. It is possible for the user to add or remove models.
    LinearRegression
    LogisticRegression
    RidgeRegression
    RidgeClassifier
    LassoRegression
    SGDRegressor
    SGDClassifier
    GaussianNB
    MultinomialNB
    ElasticNet
    BayesianRidge
    KNeighborsRegressor
    KNeighborsClassifier
    RadiusNeighborsRegressor
    RadiusNeighborsClassifier
    GaussianProcessRegressor
    GaussianProcessClassifier
    DecisionTreeRegressor
    DecisionTreeClassifier
    RandomForestRegressor
    RandomForestClassifier
    ExtraTreesRegressor
    ExtraTreesClassifier
    GradientBoostingRegressor
    GradientBoostingClassifier
            
The function automatically tunes the hyperparameters by randomized search through the hyperparameter grid provided in the CandidateModels function. The parameter grid can be edited to fit the needs of the project by the end-user.
Currently, r2 is used for regression models and accuracy is used for the classification, different metrics are being looked into.

Contact

If you encounter any issues or errors while implementing theOneMLFunction, please feel free to contact me at [email protected]. I'll do my best to help you out!

Please note that theOneMLFunction is still in active development, and the following features are in the planning or implementation phase:

ClassBalancerChecker: Checks if there is class imbalance and implements class_weights or under/oversampling as necessary
More detailed output: Implementing metrics functions for proper validation metrics, and feature importance / coefficient plots
Multi-Label Classification: Adding support for multi-label classification tasks
Multi-Output Regression: Adding support for multi-output regression task

Thank you for your understanding, and I welcome any feedback or suggestions for improvements!