/SMAI-Spring2019-Project

​SVM implementation​ ​with​ ​different ​ ​kernel(s)​ ​from​ ​scratch.

Primary LanguageJupyter NotebookMIT LicenseMIT

SMAI Spring2019 Project

Implement MULTICLASS SVM WITH DIFFERENT KERNELS FROM SCRATCH

  • Team No. : 44
  • Project ID : 26

Team Members :

  1. 2018801010 - Karnati Venkata Kartheek
  2. 2018900061 - Shashikant Ghangare

Install Dependencies

  • Install virtualenv, virtualenvwrapper
    sudo pip3 install virtualenv virtualenvwrapper

  • Create a SMCSVM Virtual environment and activate it
    mkvirtualenv SMCSVM
    workon SMCSVM

  • Install required Python Packages
    pip3 install -r requiremnts.txt

Usage

  • Open MCSVM.ipynb for SMCSVM algorithm.

  • To create SMCSVM object use:
    clf = SMCSVM()

  • Pass training data and trainng labels to fit() func'tion to train the classifier.
    clf.fit(train_X_data, train_y_label)

  • You can also pass folowing parameters:

    1. C, default_value, C=10 - Penalizing factor for Slack
    2. kernel, default value, kernel='rbf', can also take - 'linear', 'polynomial' for utilizing kernel tricks on Non-Linear data.
    3. sigma, default_value=1.0, required for 'RBF' kernel.
    4. degree, default_value=1, degree of polynomial function used in 'polynomial' kernel.

  • To predict on testing data use:
    clf.predict(test_X_data)

  • The algorithm uses K-fold cross validation as a performance metric.

Run tests

  • To run tests, run the run_tests.ipynb file in tests directory

References