A Custom Implementation of the Naive Bayes Gaussian Algorithm

Description

This package is an application of the Naive Bayes Gaussian algorithm that is commonly used to classify objects whose attributes are continuous data.

Please click here to read my brief introduction to Bayesian statistics and a use case of my custom implementation of the Naive Bayes Gaussian algorithm.

Dependencies

This package uses the following libraries.

• Python 3.8
• pandas
• numpy
• plotly

Installing and Executing program

1. Pip install the package

   pip install NaiveBayesGauss
   

2. Import the model

   from NaiveBayes import NaiveBayesGauss
   

3. Instantiate the model

   model = NaiveBayesGauss()
   

4. Fit the training data

   model.fit(X_train, Y_train)
   

5. Use the fitted model to predict a class using a single observation of attributes

   model.predict(X_target.iloc[10], use_normalizer=True)
   

6. Obtain the preceeding prediction's complete prediction probability values

   model.predict_prob
   

7. Calculate the model's prediction accuracy on fitted data and plot a confusion matrix

   model.predict_accuracy(X_target, Y_target, user_normalizer=True, confusion_matrix=True)
   

8. Calculate the model's prediction accuracy on fitted data and plot the results on a heat map

   model.plot_heatmap(X_train, Y_train, attributes['SepalLengthCm', 'PetalLengthCm'], predict_label='Iris-setosa', h=0.1)
   

Authors

Ilya Novak @NovakIlya

Version History

• 0.1
  • Initial Release

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Acknowledgments

• Lectures by Victor Lavrenko were very helpful in understanding the Naive Bayes Gauss algorithm
• The Iris dataset used to validate the model accuracy was obtained from UC Irvine Machine Learning Repository
• The model accuracy was tested against the sklearn implementation
• The code for the heat map was obtained from Arthur Tok