ryanbae89/mklearn

My implementation of scikit-learn, mykit-learn. Code release for DATA 558 spring 2018.

Mykit-Learn

My implementation of scikit-learn, mykit-learn. Mklearn is my implementation of Linear Support Vector Machine model. It is implemented similarly to scikit-learn using classes and fit functions, etc.

Usage

Make sure Python has the correct path to the modules, and simply import as follows:

import mklearn
import multiclass

mklearn.myLinearSVC(...)
multiclass.Multiclass(...)

Implementation

Linear Support Vector Classifier

mklearn.py contains the implementation of linear support vector classifier. The instance of class myLinearSVC can fit a single binary classifier model using linear SVC. The class is structured in a similar way to scikit-learn's convention, where the class is declared with hyperparameters and .fit() method is called to train the model.

Multiclass Classification

Multiclass classification is supported by multiclass.py file. It contains Multiclass object and several prediction functions that wrap around myLinearSVC. Multiclass can perform both one-vs-one and one-vs-rest multiclass classification through the argument multiclass.

The module also supports model specific cross-validation while running multiclass classification. Simply pass in a list of regularization penalties as Cs and set k to the number of desired folds for cross-validation.

Finally, Multiclass class supports limited multiprocessing to speed up training. Currently, it is only available for the prediction portion only. You can pass in number of threads into n_threads argument. The multiprocessing capabilities for training portion is in development.

Dependencies

The following dependencies are needed to install and run mklearn.

numpy, pandas, matplotlib, scipy, sklearn

Currently mkleran is only supported in Python 3.

Tutorial

There are 3 tutorials to showcase how to use mklearn.

• mklearn_tutorial_real_dataset.ipynb: contains mklearn multiclass example on real-world dataset.

• mklearn_tutorial_simulated_dataset.ipynb: contains mklearn multiclass example on simulated dataset using sklearn.

• tutorial_sklearn_comparison.ipynb: contains performance comparisons with similar implementations from sklearn.

Take a look at the mentioned notebooks for details.

Support/Contact

Author: Ryan Bae

Initial Commit Date: 6/02/2018

This is my official code release for DATA 558 Spring 2018 at University of Washington's MS Data Science program.