Skip to content

Commit

Permalink
Script for CajaMar experiments
Browse files Browse the repository at this point in the history
  • Loading branch information
@anamf
anamf committed May 7, 2016
1 parent 51c8900 commit 7b98ef4
Show file tree
Hide file tree
Showing 4 changed files with 352 additions and 3 deletions.
Binary file removed extensions/nips2016/doc-Experiments/preliminaryExperiments/OneGaussian.xlsx
View file
Binary file not shown.
351 changes: 351 additions & 0 deletions extensions/nips2016/src/main/java/eu/amidst/nips2016/CajaMarExperiments.java
View file
@@ -0,0 +1,351 @@
package eu.amidst.nips2016;

import eu.amidst.core.datastream.Attributes;
import eu.amidst.core.datastream.DataInstance;
import eu.amidst.core.datastream.DataOnMemory;
import eu.amidst.core.datastream.DataStream;
import eu.amidst.core.distribution.Normal;
import eu.amidst.core.distribution.Normal_MultinomialParents;
import eu.amidst.core.inference.messagepassing.VMP;
import eu.amidst.core.io.DataStreamLoader;
import eu.amidst.core.learning.parametric.ParallelMaximumLikelihood;
import eu.amidst.core.learning.parametric.bayesian.DriftSVB;
import eu.amidst.core.learning.parametric.bayesian.PopulationVI;
import eu.amidst.core.learning.parametric.bayesian.SVB;
import eu.amidst.core.learning.parametric.bayesian.StochasticVI;
import eu.amidst.core.models.BayesianNetwork;
import eu.amidst.core.models.DAG;
import eu.amidst.core.variables.Variable;
import eu.amidst.core.variables.Variables;

import java.io.PrintWriter;
import java.util.ArrayList;
import java.util.List;

/**
* Created by ana@cs.aau.dk on 04/05/16.
*/
public class CajaMarExperiments {

public static boolean classVarInModel = true;
public static boolean linkHidden = true;

public static int[] batchSize = {1000};
public static int[] memoryPopulationVI = {1000};
public static double[] learningRate = {0.8};
public static double[] deltaValue = {100};

public static SVB svb;
public static DriftSVB driftSVB;
public static PopulationVI populationVI;
public static StochasticVI stochasticVI;
public static ParallelMaximumLikelihood ml;
public static ParallelMaximumLikelihood mlPerBatch;

public static PrintWriter writerPredLL;
public static PrintWriter writerLambda ;
public static PrintWriter writerMean;
public static PrintWriter writerGamma;

public static DAG dag;

public static int numIter = 84;

public static int iter = 0;

public static int maxIterVI = 100;
public static double thresholdVI = 0.001;

public static void initSVBLearners(int batchSize, double deltaValue) {
driftSVB = new DriftSVB();
driftSVB.setWindowsSize(batchSize);
driftSVB.setSeed(0);
driftSVB.setDelta(deltaValue);
driftSVB.setOutput(true);
VMP vmp = driftSVB.getPlateuStructure().getVMP();
//vmp.setOutput(true);
vmp.setTestELBO(true);
vmp.setMaxIter(maxIterVI);
vmp.setThreshold(thresholdVI);
driftSVB.setDAG(dag);
driftSVB.initLearning();

svb = new SVB();
svb.setWindowsSize(batchSize);
svb.setSeed(0);
svb.setOutput(true);
vmp = svb.getPlateuStructure().getVMP();
vmp.setOutput(true);
vmp.setTestELBO(true);
vmp.setMaxIter(maxIterVI);
vmp.setThreshold(thresholdVI);
svb.setDAG(dag);
svb.initLearning();

}

public static void initVILearners(int batchSize, int memoryPopulationVI, double learningRate){

populationVI = new PopulationVI();
populationVI.setMemorySize(memoryPopulationVI);
populationVI.setBatchSize(batchSize);
populationVI.setSeed(0);
populationVI.setOutput(true);
populationVI.setLearningFactor(learningRate);
VMP vmp = svb.getPlateuStructure().getVMP();
vmp.setOutput(true);
vmp.setTestELBO(true);
vmp.setMaxIter(maxIterVI);
vmp.setThreshold(thresholdVI);
populationVI.setDAG(dag);
populationVI.initLearning();

stochasticVI = new StochasticVI();
stochasticVI.setDataSetSize(numIter*batchSize);
stochasticVI.setBatchSize(batchSize);
stochasticVI.setSeed(0);
stochasticVI.setOutput(true);
stochasticVI.setLearningFactor(learningRate);
vmp = svb.getPlateuStructure().getVMP();
//vmp.setOutput(true);
vmp.setTestELBO(true);
vmp.setMaxIter(maxIterVI);
vmp.setThreshold(thresholdVI);
stochasticVI.setDAG(dag);
stochasticVI.initLearning();

maximumLikelihoodInit();
}


private static void maximumLikelihoodInit(){
ml = new ParallelMaximumLikelihood();
ml.setParallelMode(true);
ml.setDAG(dag);
ml.initLearning();

mlPerBatch = new ParallelMaximumLikelihood();
mlPerBatch.setParallelMode(true);
mlPerBatch.setDAG(dag);
mlPerBatch.initLearning();
}

public static void printOutput() throws Exception{

BayesianNetwork bnML = ml.getLearntBayesianNetwork();
BayesianNetwork bnSVB = svb.getLearntBayesianNetwork();
BayesianNetwork bnDriftSVB = driftSVB.getLearntBayesianNetwork();
BayesianNetwork bnStochasticVI = stochasticVI.getLearntBayesianNetwork();
BayesianNetwork bnPopulationVI = populationVI.getLearntBayesianNetwork();
BayesianNetwork bnMLPerBatch = mlPerBatch.getLearntBayesianNetwork();


double[] meanML=new double[2], meanSVB=new double[2], meanDriftSVB=new double[2], meanStochasticVI=new double[2],
meanPopulationVI=new double[2], realMean=new double[2];

if(!classVarInModel) {
meanML[0] = ((Normal) bnML.getConditionalDistribution(dag.getVariables().getVariableByName("VAR01"))).getMean();
meanSVB[0] = ((Normal) bnSVB.getConditionalDistribution(dag.getVariables().getVariableByName("VAR01"))).getMean();
meanDriftSVB[0] = ((Normal) bnDriftSVB.getConditionalDistribution(dag.getVariables().getVariableByName("VAR01"))).getMean();
meanStochasticVI[0] = ((Normal) bnStochasticVI.getConditionalDistribution(dag.getVariables().getVariableByName("VAR01"))).getMean();
meanPopulationVI[0] = ((Normal) bnPopulationVI.getConditionalDistribution(dag.getVariables().getVariableByName("VAR01"))).getMean();
realMean[0] = ((Normal)bnMLPerBatch.getConditionalDistribution(dag.getVariables().getVariableByName("VAR01"))).getMean();
}else{
for (int i = 0; i < 2; i++) {
meanML[i] = ((Normal_MultinomialParents) bnML.getConditionalDistribution(dag.getVariables().getVariableByName("VAR01"))).getNormal(i).getMean();
meanSVB[i] = ((Normal_MultinomialParents) bnSVB.getConditionalDistribution(dag.getVariables().getVariableByName("VAR01"))).getNormal(i).getMean();
meanDriftSVB[i] = ((Normal_MultinomialParents) bnDriftSVB.getConditionalDistribution(dag.getVariables().getVariableByName("VAR01"))).getNormal(i).getMean();
meanStochasticVI[i] = ((Normal_MultinomialParents) bnStochasticVI.getConditionalDistribution(dag.getVariables().getVariableByName("VAR01"))).getNormal(i).getMean();
meanPopulationVI[i] = ((Normal_MultinomialParents) bnPopulationVI.getConditionalDistribution(dag.getVariables().getVariableByName("VAR01"))).getNormal(i).getMean();
realMean[i] = ((Normal_MultinomialParents)bnMLPerBatch.getConditionalDistribution(dag.getVariables().getVariableByName("VAR01"))).getNormal(i).getMean();
}
}

String means = "";
for (int i = 0; i < 2; i++) {
if(i!=0)
means += "\t";
means += realMean[i]+"\t"+meanML[i]+"\t"+meanSVB[i]+"\t"+meanDriftSVB[i]+"\t"+meanStochasticVI[i]+"\t"+meanPopulationVI[i];
}


writerMean.println(means);


writerLambda.println(driftSVB.getLambdaValue());

}

public static void printCounts() throws Exception{

double[] outputs = new double[4];
outputs[0] = svb.getPlateuStructure().getNonReplictedNodes().findFirst().get().getQDist().getNaturalParameters().get(0);
outputs[1] = driftSVB.getPlateuStructure().getNonReplictedNodes().findFirst().get().getQDist().getNaturalParameters().get(0);
outputs[2] = stochasticVI.getSVB().getPlateuStructure().getNonReplictedNodes().findFirst().get().getQDist().getNaturalParameters().get(0);
outputs[3] = populationVI.getSVB().getPlateuStructure().getNonReplictedNodes().findFirst().get().getQDist().getNaturalParameters().get(0);
writerGamma.println(outputs[0]+"\t"+outputs[1]+"\t"+outputs[2]+"\t"+outputs[3]);



}

public static void printPredLL(DataOnMemory<DataInstance> batch) throws Exception{

double[] outputs = new double[4];
outputs[0] += svb.predictedLogLikelihood(batch);
outputs[1] += driftSVB.predictedLogLikelihood(batch);
outputs[2] += stochasticVI.predictedLogLikelihood(batch);
outputs[3] += populationVI.predictedLogLikelihood(batch);
writerPredLL.println(outputs[0]/iter+"\t"+outputs[1]/iter+"\t"+outputs[2]/iter+"\t"+outputs[3]/iter);


}

public static DAG createDAG(Attributes attributes, int nlocals){
// Create a Variables object from the attributes of the input data stream.
Variables variables = new Variables(attributes);

// Define the class variable.
Variable classVar = variables.getVariableByName("DEFAULTING");

// Define the local hidden variables.
List<Variable> localHiddenVars = new ArrayList<>();
for (int i = 0; i < nlocals; i++) {
localHiddenVars.add(variables.newGaussianVariable("Hidden_"+i));
}


// Create an empty DAG object with the defined variables.
DAG dag = new DAG(variables);

// Link the class as parent of all attributes
if(classVarInModel) {
dag.getParentSets()
.stream()
.filter(w -> w.getMainVar() != classVar)
.filter(w -> !w.getMainVar().getName().startsWith("Hidden"))
.forEach(w -> w.addParent(classVar));
}

// Link the local hidden as parent of all predictive attributes
for (Variable localHiddenVar : localHiddenVars) {
dag.getParentSets()
.stream()
.filter(w -> w.getMainVar() != classVar)
.filter(w -> !w.getMainVar().getName().startsWith("Hidden"))
.forEach(w -> w.addParent(localHiddenVar));
}


// Connect local hidden variables with each other
if(linkHidden) {
for (int i = 0; i < localHiddenVars.size() - 1; i++) {
for (int j = i + 1; j < localHiddenVars.size(); j++) {
dag.getParentSet(localHiddenVars.get(i)).addParent(localHiddenVars.get(j));
}

}
}




// Show the new dynamic DAG structure
System.out.println(dag.toString());

return dag;
}

public static void main(String[] args) throws Exception{

//int[] peakMonths = {2, 8, 14, 20, 26, 32, 38, 44, 47, 50, 53, 56, 59, 62, 65, 68, 71, 74, 77, 80, 83};

String path="/Users/ana/Documents/Amidst-MyFiles/CajaMar/dataWeka/dataWeka";
String outputPath = "extensions/nips2016/doc-Experiments/preliminaryExperiments/";

for (int i = 0; i < batchSize.length; i++) {

for (int j = 0; j < deltaValue.length; j++) {

for (int k = 0; k < memoryPopulationVI.length; k++) {

for (int l = 0; l < learningRate.length; l++) {


for (int m = 0; i < numIter; m++) {

int currentMonth = m;

//if (IntStream.of(peakMonths).anyMatch(x -> x == currentMonth))
// continue;

DataStream<DataInstance> dataMonthi = DataStreamLoader.openFromFile(path + currentMonth + ".arff");

dag = createDAG(dataMonthi.getAttributes(), 5);

/**
* Define Learning VI techniques
*/
initSVBLearners(batchSize[i], deltaValue[j]);
initVILearners(batchSize[i], memoryPopulationVI[k], learningRate[l]);

/**
* Update with all different learning techniques
*/
int batchCount = 0;
for (DataOnMemory<DataInstance> batch : dataMonthi.iterableOverBatches(batchSize[i])) {

System.out.println("--------------------------------- MONTH "+currentMonth+"/"+batchCount+ " --------------------------");
driftSVB.updateModelWithConceptDrift(batch);
svb.updateModel(batch);
populationVI.updateModel(batch);
stochasticVI.updateModel(batch);

/* Learn maximum likelihood to get the real means*/
ml.updateModel(batch);
mlPerBatch.initLearning();
mlPerBatch.updateModel(batch);

if(i>0) {
printPredLL(batch);
printCounts();
}
batchCount++;
}

//We print the output
/**
* Output files for predLL, lambda, mean, population size
*/
writerPredLL = new PrintWriter(outputPath + "CajaMar_predLL_" + "bs" + batchSize[i] + "_delta" +
deltaValue[j]+ "_mem" + memoryPopulationVI[k] + "_lr" + learningRate[l] +
".txt", "UTF-8");
writerLambda = new PrintWriter(outputPath + "CajaMar_lamda_" + "_bs" + batchSize[i] + "_delta" +
deltaValue[j]+ "_mem" + memoryPopulationVI[k] + "_lr" + learningRate[l] +
".txt", "UTF-8");
writerMean = new PrintWriter(outputPath + "CajaMar_mean_" + "bs" + batchSize[i] + "_delta" +
deltaValue[j]+ "_mem" + memoryPopulationVI[k] + "_lr" + learningRate[l] +
".txt", "UTF-8");
writerGamma = new PrintWriter(outputPath + "CajaMar_gamma_" + "bs" + batchSize[i] + "_delta" +
deltaValue[j]+ "_mem" + memoryPopulationVI[k] + "_lr" + learningRate[l] +
".txt", "UTF-8");

/**
* Outputs: lambda, mean, population size
*/
printOutput();

/**
* Close all output files
*/
writerPredLL.close();
writerLambda.close();
writerMean.close();
writerGamma.close();
}
}
}
}
}
}
}
2 changes: 0 additions & 2 deletions extensions/nips2016/src/main/java/eu/amidst/nips2016/PreliminaryExperiments.java
View file
Expand Up @@ -139,8 +139,6 @@ public static void printPredLL(DataOnMemory<DataInstance> batch) throws Exceptio
outputs[2] += stochasticVI.predictedLogLikelihood(batch);
outputs[3] += populationVI.predictedLogLikelihood(batch);
writerPredLL.println(outputs[0]/iter+"\t"+outputs[1]/iter+"\t"+outputs[2]/iter+"\t"+outputs[3]/iter);


}

public static void printOutput() throws Exception{
Expand Down
2 changes: 1 addition & 1 deletion run.sh
View file
@@ -1,3 +1,3 @@
#!/bin/bash

java -Xmx5000M -Djava.library.path="./huginlink/huginlib/" -cp "./huginlink/huginlib/*:./core/target/*:./examples/target/*:./moalink/target/*:./huginlink/target/*:./flinklink/target/*:./extensions/uai2016/target/*" $@
java -Xmx5000M -Djava.library.path="./huginlink/huginlib/" -cp "./huginlink/huginlib/*:./core/target/*:./examples/target/*:./moalink/target/*:./huginlink/target/*:./flinklink/target/*:./extensions/uai2016/target/*:./extensions/nips2016/target/*" $@

0 comments on commit 7b98ef4

Please sign in to comment.