SVM-Devnagari-Handwritten-Character-Recognition

Solving the Character recognition problem as an SVM optimization problem using CVXOPT

##Support Vector Machine

Name

svm - Run the executable program for SVM

Synopsis Part a./svm <part> <tr> <ts> <out> <c_value> Part b./svm <part> <tr> <ts> <out> <c_value> <gamma> Part c./svm <part> <tr> <ts> <out> <c_value> <gamma>

Description

This program will train svm model using given code on train data, make predictions on test data and write final predictions in given output file.

Options

-part
Part i.e. a/b/c.
-tr
File containing training data in csv format where 1st entry is the target
-ts
File containing test data in csv format where 1st entry is the target
-out
Output file for predictions. One value in each line. -c_value
C is a regularization parameter that controls the trade-off between maximizing the margin and minimizing the training error.
-gamma
Bandwidth parameter for RBF kernel

Example

./svm a DHC_train.csv DHC_test.csv output 10 ./svm b DHC_train.csv DHC_test.csv output 10 0.01 ./svm c DHC_train.csv DHC_test.csv output 10 0.01

Parts

• Part A

  • Expressed the SVM dual problem using a linear kernel
  • Soft Margin formulation

• Part B

  • Solved the dual SVM Problem using a RBF Kernel(Gaussian Kernel)

• Part C

  • Implemented PCA algorithm using the SVD formulation.
  • Applied SVM with RBF Kernel on the projected data

Data

• DHC_train.csv: Train data
• DHC_test.csv: Test data