AI Practice 05 / 07

Sang-Woo Lee

© 2013, SNU CSE Biointelligence Lab., http://bi.snu.ac.kr 2

1. Usage of SVM and Decision Tree in Weka2. Amplification about Final Project Spec3. SVM – State of the Art in Classification

4. Commentary for Results on Mid-Term Project

5. Useful Technique for Final Project6. Decision Tree

Contents

USAGE OF SVM & DECISION TREES IN WEKA

© 2013, SNU CSE Biointelligence Lab., http://bi.snu.ac.kr 4

Neural Networks

• MLP (Multilayer Perceptron)

– In Weka, Classifiers-functions-MultilayerPerceptron

© 2013, SNU CSE Biointelligence Lab., http://bi.snu.ac.kr 5

Support Vector Machines

SMO (sequential minimal optimization) for training SVM In Weka, classifiers-functions-SMO

© 2013, SNU CSE Biointelligence Lab., http://bi.snu.ac.kr 6

Decision Trees

• J48 (Java implementation of C4.5)

– In Weka, classifiers-trees-J48

© 2013, SNU CSE Biointelligence Lab., http://bi.snu.ac.kr 7

Some Notes on the Parameter Setting

• Parameter Setting = Car Tuning– need much experience or many times of trial – you may get worse results if you are unlucky

• Multilayer Perceptron (MLP)– Main parameters for learning: hiddenLayers, learningRate, momentum,

trainingTime (epoch), seed

• SMO (SVM)– Main parameters: c (complexity parameter), kernel, kernel parameters

• J48– Main parameters: unpruned, numFolds, minNumObj– Many parameters are for controlling the size of the result tree, i.e. confi-

denceFactor, pruning

© 2013, SNU CSE Biointelligence Lab., http://bi.snu.ac.kr 8

Using SVM in Weka

© 2013, SNU CSE Biointelligence Lab., http://bi.snu.ac.kr 9

Using DT in Weka

SPEC OF FINAL PROJECT

Final Project (Mandatory)• Extension experiments of midterm project• Methods (2 mandatory algorithm)– Must Use Weka

• MLP (used in midterm)• SVM or Decision Tree (mandatory)

• If you just use 2 mandatory algorithm (MLP and one more) and make rea-sonable result, you get default point.

Final Project (Optional)

• Methods (1 more optional algorithm)– Must Use Weka

• One arbitrary algorithm in weka

• Optional issue– Proper arbitrary algorithm for data– Proper preprocessing for data– Dataset used in midterm project

• If you are interested in this project, you could try optional issue and you get bonus point.

Final Report

• Final report and presentation– Submit report– About: Design your problem– Include

• Introduction• Problem definition• Dataset• Preprocess (optional)• Methods• Experimental results and comparison• Discussion

• Due date: 2013. 06. 10. PM 23:59• Submit by email (slee@bi.snu.ac.kr)

© 2013, SNU CSE Biointelligence Lab., http://bi.snu.ac.kr 14

Must Preprocess be done with weka? No. You can use anything for preprocess But, there is some preprocess tool in weka.

Isn’t it preferred to use Extension of Mid-term Dataset? It is more preferred to use Extension of Mid-

term Dataset.

If you do more whatever interesting for project, you will get more point!

But, it is ok just to satisfy mandatory one.

Q&A

© 2013, SNU CSE Biointelligence Lab., http://bi.snu.ac.kr 15

C.-W. Hsu, C.-C. Chang, C.-J. Lin, “A Practical Guide to Sup-port Vector Classification”

Some Good Materials