Digit Classification Using Support Vector Machine (SVM)

In this project, I implement digit classification using LibSVM libarary.

Datasets

• The MNIST database of handwritten digits: http://yann.lecun.com/exdb/mnist/

Requirements

• libsvm: https://github.com/cjlin1/libsvm
• scipy.io from scipy
• numpy
• matplotlib

Testing Environment

• sysname: Linux
• release: 4.15.0-45-generic
• machine: x86_64
• python: 3.6.7

Results

• Linear SVM

• RBF SVM (Radial Basis Function)

How to reproduce the results

1. Clone the libsvm repository:

git clone https://github.com/cjlin1/libsvm.git

2. Clone my repository:

git clone https://github.com/lychengr3x/Digit-Classification-Using-SVM.git

3. Move all my codes into directory python of libsvm:

cp -r Digit-Classification-Using-SVM/src/* libsvm/python/

4. Follow instructions of README in the libsvm. For linux OS, go to the directory libsvm/python:

cd libsvm/python
make

• Note: If it does not work, go to the directory libsvm:

cd ..
make

5. Now, you can execute the run_linear_svm.py and run_rbf_svm.py:

python run_linear_svm.py
python run_rbf_svm.py