https://github.com/brade31919/sequential-minimal-optimization

This is a Matlab implementation of SMO based on the original work from JC Platt
https://github.com/brade31919/sequential-minimal-optimization

Last synced: about 11 hours ago
JSON representation

This is a Matlab implementation of SMO based on the original work from JC Platt

• Host: GitHub
• URL: https://github.com/brade31919/sequential-minimal-optimization
• Owner: brade31919
• Created: 2017-04-04T18:59:46.000Z (over 7 years ago)
• Default Branch: master
• Last Pushed: 2015-09-19T18:27:59.000Z (about 9 years ago)
• Last Synced: 2023-03-10T14:42:39.276Z (over 1 year ago)
• Language: Matlab
• Size: 141 KB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Sequential-minimal-optimization

=========================

Author: Ehsan Zahedinejad [email protected]

=========================

Title:

=========================

Sequential-minimal-optimization. A matlab implementation of Sequential-minimal-optimization based on the original work from

John C. Platt 2000. See publication for more detail: http://research.microsoft.com/pubs/68391/smo-book.pdf

Benchmarks:

=========================

I have tested the code on UCI Ionosphere Data Set with a traning error less than 0.3% and confidence of more than 96% on the validation data. The training size was 200 and the validation size was 151.

Function description:

============================================

The main function is SMO(X,Y,eps,tol,type,ul,Pdata,Sigma).

Input arguments for SMO function

===========================================

1. X= The feature matrix (2d array)

2. Y= Target matrix (1d array)

3. eps= Convergence criteria--show what should the difference between lagranage multiplier between two consecutive iteration has to be to exit. Example eps=0.001 (scalar)

4. tol= Distance within that the lagrange multiplier will be mapped to zero or upper limit 'C'. Best value could be:0.001 (scalar)

5. type= What Kernel function one likes to use. Currently Linear 'L' or Gaussian function are supported 'G' (char)

6. ul= upper and lower bound for lagrange multipliers. Example [0, 1]. If the upper is infinity write [0 Inf] (1d array) 

7. Pdata= Percenateg of data being used for training data. The rest will be used for validation (scalar in terms of percentage)

8. Sigma= Gaussian variance. If liner kernel write zero. 

Output arguments for SMO function

===========================================

1. [alpha,b,TS]=SMO(X,Y,eps,tol,type,ul,Pdata,Sigma).

2. alpha= array of lagrange multipliers (1d array)

3. b= threshold value (scalar)

4. TS= training size

5. Training error= SMO prints out the error when a run is complete

6. Confidence of hypothesis= SMO prints out the confidence when a run is complete