Machine Learning with Weka and Spring examples

Code url: https://github.com/dmmiller612/Machine_Learning_Spring_Weka

Instructions

If wanting to run the server locally, instead of just using the Weka models located in /src/main/resources/models, there are a couple of dependencies needed: Maven and Java.

1. This uses java 1.7, but should work with 1.8 as well. For the JRE, sudo apt-get install default-jre . For the jdk, sudo apt-get install default-jdk.

2. This uses Maven 3.x . To Install maven 3, use sudo apt-get install maven.

3. Go to the root of the assignment code repository and type: mvn clean package into the command line. Then type java -jar target/derek-assignment-1-0.1.0.jar. Once running the jar, all of optimal models will start to run against the test datasets of both the Car Evaluation and Census dataset. This is here just to make it easier to visualize, so that you do not have to use the rest api. If you want to use the rest api, see documentation below.

4. IF three does not work, it is because the plugin did not properly install. Running this command should do the trick inside of the students-filters-master "mvn install:install-file -Dfile=filters-0.0.1-SNAPSHOT.jar -DgroupId=filters -DartifactId=filters -Dversion=0.0.1-SNAPSHOT -Dpackaging=jar"

Navigating the Source Code

src/main/java/com/derek/ml/controllers

contains the rest endpoints.

src/main/java/com/derek/ml/services

contains all of the logic and configuration of weka models. ClusterService -> k-means and EM, FeatureReductionService -> ICA, PCA, RP, CFS, KNNService -> KNN, NeuralNetworkService -> Neural Network, DecisionTreeService->Decision Trees (boosted an unboosted), SVMService->SVM

src/main/java/com/derek/ml/models

DTO passing layers

Navigating the Resources

src/main/resources/csv

Contains all of the initial csv files used (Arffs are only used for the models, however)

src/main/resources/arffs

Contains all of the arffs used. car_train.arff and car_test.arff are the training and testing instances for the car evaluation dataset. census.arff and censusTest.arff are the training and testing instances for the Census dataset.

src/main/resources/models

Contains several models used for the supervised learning analysis. If you don’t want to run the code locally, you can just use these models against the training and test arffs listed above.

Using the Rest API (Optional)

I thought I would just add this to show the code I used for experimentation with Weka. I used the api, so that I could do multiple concurrent requests.

Universal Query parameters: fileName : {Car, Census, CarBin, CensusBin}, testType : {CrossValidation, TestData, Train}

Cluster

Endpoints: /kMeans and /em Query Params => clusters : int, distances : {Euclidean, Manhatten}, iterations: int, featureSelection: {ICA, PCA, RP, CFS};

Feature Reduction

Endpoints: /featureReduction/pca /featureReduction/ica /featureReduction/rp /featureReduction/cfs

Decision Trees

Endpoint: /decisiontree

Query Params => minNumObj : int, boost : boolean, confidence : String, treeType : {ID3, J48}

Example Requests: http://localhost:8080/decisiontree?fileName=Car&testType=TestData&minNumObj=2&confidence=.25 http://localhost:8080/decisiontree?fileName=Census&testType=CrossValidation&minNumObj=2&confidence=.25 http://localhost:8080/decisiontree?fileName=Car&testType=CrossValidation&minNumObj=2&confidence=.25&boost=true //with boosting

Using incremental testing example: http://localhost:8080/decisiontree/test?fileName=Car&testType=TestData&minNumObj=2&confidence=.25&boost=true

KNN

Endpoint: /knn

Query Params => k : int, boost : boolean, treeTypes {BallTree, CoverTree, Linear}, useFeatureSelection : boolean (applies only to Census file)

Examples : http://localhost:8080/knn?fileName=Car&testType=TestData&k=3 http://localhost:8080/knn?fileName=Census&testType=TestData&k=5&featureSelection=true http://localhost:8080/knn?fileName=Census&testType=TestData&k=5&boost=true

Using incremental testing example: http://localhost:8080/knn/test?fileName=Census&testType=TestData&k=5

ANN

Endpoint: /neuralnetwork

Query Params => hiddenLayers : int, epochRate : int, featureSelection : boolean (applies only to Census file)

Examples: http://localhost:8080/neuralnetwork?fileName=Car&testType=TestData&hiddenLayers=10&epochRate=500 http://localhost:8080/neuralnetwork?fileName=Census&testType=TestData&hiddenLayers=5&epochRate=500&featureSelection=true

Using incremental testing example: http://localhost:8080/neuralnetwork/test?fileName=Car&testType=TestData&hiddenLayers=10&epochRate=500

SVM

Endpoint: /svm

Query Params => kernelType : {Polynomial, RBF, Sigmoid, Linear}

Examples: http://localhost:8080/svm?fileName=Car&testType=TestData&kernelType=Polynomial http://localhost:8080/svm?fileName=Census&testType=TestData&kernelType=RBF http://localhost:8080/svm?fileName=Census&testType=TestData&kernelType=Sigmoid http://localhost:8080/svm?fileName=Census&testType=TestData&kernelType=Linear

MODELS

The model names contain the parameters that were used, fileName, and algorithm name.

Decision Tree Naming Convention: decisionTree + minNumObj + Boosted + confidence + fileName + .model Example: decisionTree-minNumObj=100-Boosted=false-C=0.25-file=Census.model

KNN Naming Convention: KNearestNeighbor + k + fileName + .model Example : KNearestNeighbor-k=20-fileName=Car.model

ANN Naming Convention: ANN + hiddenLayers + epochRate + FileName + (Optional) featureSelection + .model Example : ANN-hiddenLayers=10-epochRate=250-FileName=Census.model

SVM Naming Convention: SVM + kernelType + FileName + .model Example : SVM-KernelType=Linear-FileName=Car.model