Examples with LargeScaleSimulation edu.cmu.tetrad.sem.LargeScaleSimulation used on opensource projects

Example 16 with LargeScaleSimulation

use of edu.cmu.tetrad.sem.LargeScaleSimulation in project tetrad by cmu-phil.

the class PerformanceTests method testFges.

public void testFges(int numVars, double edgeFactor, int numCases, double penaltyDiscount) {
    init(new File("long.fges." + numVars + "." + edgeFactor + "." + penaltyDiscount + ".txt"), "Tests performance of the FGES algorithm");
    long time1 = System.currentTimeMillis();
    Graph dag = makeDag(numVars, edgeFactor);
    System.out.println("Graph done");
    out.println("Graph done");
    System.out.println("Starting simulation");
    LargeScaleSimulation simulator = new LargeScaleSimulation(dag);
    simulator.setOut(out);
    DataSet data = simulator.simulateDataFisher(numCases);
    System.out.println("Finishing simulation");
    long time2 = System.currentTimeMillis();
    out.println("Elapsed (simulating the data): " + (time2 - time1) + " ms");
    System.out.println("Making covariance matrix");
    ICovarianceMatrix cov = new CovarianceMatrixOnTheFly(data);
    System.out.println("Covariance matrix done");
    long time3 = System.currentTimeMillis();
    out.println("Elapsed (calculating cov): " + (time3 - time2) + " ms");
    SemBicScore semBicScore = new SemBicScore(cov);
    semBicScore.setPenaltyDiscount(penaltyDiscount);
    Fges pcStable = new Fges(semBicScore);
    Graph estPattern = pcStable.search();
    // out.println(estPattern);
    long time4 = System.currentTimeMillis();
    // out.println("# Vars = " + numVars);
    // out.println("# Edges = " + (int) (numVars * edgeFactor));
    out.println("# Cases = " + numCases);
    out.println("penalty discount = " + penaltyDiscount);
    out.println("Elapsed (simulating the data): " + (time2 - time1) + " ms");
    out.println("Elapsed (calculating cov): " + (time3 - time2) + " ms");
    out.println("Elapsed (running FGES) " + (time4 - time3) + " ms");
    out.println("Total elapsed (cov + FGES) " + (time4 - time2) + " ms");
    final Graph truePattern = SearchGraphUtils.patternForDag(dag);
    System.out.println("# edges in true pattern = " + truePattern.getNumEdges());
    System.out.println("# edges in est pattern = " + estPattern.getNumEdges());
    SearchGraphUtils.graphComparison(estPattern, truePattern, out);
    out.println("seed = " + RandomUtil.getInstance().getSeed() + "L");
    out.close();
}

Example 17 with LargeScaleSimulation

use of edu.cmu.tetrad.sem.LargeScaleSimulation in project tetrad by cmu-phil.

the class PerformanceTests method testFges.

private void testFges(int numVars, double edgeFactor, int numCases, int numRuns, boolean continuous) {
    out.println(new Date());
    // RandomUtil.getInstance().setSeed(4828384343999L);
    double penaltyDiscount = 4.0;
    int maxIndegree = 5;
    boolean faithfulness = true;
    // RandomUtil.getInstance().setSeed(50304050454L);
    List<int[][]> allCounts = new ArrayList<>();
    List<double[]> comparisons = new ArrayList<>();
    List<Double> degrees = new ArrayList<>();
    List<Long> elapsedTimes = new ArrayList<>();
    if (continuous) {
        init(new File("fges.comparison.continuous" + numVars + "." + (int) (edgeFactor * numVars) + "." + numCases + "." + numRuns + ".txt"), "Num runs = " + numRuns);
        out.println("Num vars = " + numVars);
        out.println("Num edges = " + (int) (numVars * edgeFactor));
        out.println("Num cases = " + numCases);
        out.println("Penalty discount = " + penaltyDiscount);
        out.println("Depth = " + maxIndegree);
        out.println();
    } else {
        init(new File("fges.comparison.discrete" + numVars + "." + (int) (edgeFactor * numVars) + "." + numCases + "." + numRuns + ".txt"), "Num runs = " + numRuns);
        out.println("Num vars = " + numVars);
        out.println("Num edges = " + (int) (numVars * edgeFactor));
        out.println("Num cases = " + numCases);
        out.println("Sample prior = " + 1);
        out.println("Structure prior = " + 1);
        out.println("Depth = " + 1);
        out.println();
    }
    for (int run = 0; run < numRuns; run++) {
        out.println("\n\n\n******************************** RUN " + (run + 1) + " ********************************\n\n");
        System.out.println("Making dag");
        out.println(new Date());
        Graph dag = makeDag(numVars, edgeFactor);
        System.out.println(new Date());
        System.out.println("Calculating pattern for DAG");
        Graph pattern = SearchGraphUtils.patternForDag(dag);
        List<Node> vars = dag.getNodes();
        int[] tiers = new int[vars.size()];
        for (int i = 0; i < vars.size(); i++) {
            tiers[i] = i;
        }
        System.out.println("Graph done");
        long time1 = System.currentTimeMillis();
        out.println("Graph done");
        System.out.println(new Date());
        System.out.println("Starting simulation");
        Graph estPattern;
        long elapsed;
        if (continuous) {
            LargeScaleSimulation simulator = new LargeScaleSimulation(dag, vars, tiers);
            simulator.setVerbose(false);
            simulator.setOut(out);
            DataSet data = simulator.simulateDataFisher(numCases);
            System.out.println("Finishing simulation");
            System.out.println(new Date());
            long time2 = System.currentTimeMillis();
            out.println("Elapsed (simulating the data): " + (time2 - time1) + " ms");
            System.out.println(new Date());
            System.out.println("Making covariance matrix");
            long time3 = System.currentTimeMillis();
            ICovarianceMatrix cov = new CovarianceMatrixOnTheFly(data, true);
            System.out.println("Covariance matrix done");
            out.println("Elapsed (calculating cov): " + (time3 - time2) + " ms\n");
            SemBicScore score = new SemBicScore(cov);
            score.setPenaltyDiscount(penaltyDiscount);
            System.out.println(new Date());
            System.out.println("\nStarting FGES");
            long timea = System.currentTimeMillis();
            Fges fges = new Fges(score);
            // fges.setVerbose(false);
            fges.setNumPatternsToStore(0);
            fges.setOut(System.out);
            fges.setFaithfulnessAssumed(faithfulness);
            fges.setCycleBound(-1);
            long timeb = System.currentTimeMillis();
            estPattern = fges.search();
            long timec = System.currentTimeMillis();
            out.println("Time for FGES constructor " + (timeb - timea) + " ms");
            out.println("Time for FGES search " + (timec - timea) + " ms");
            out.println();
            out.flush();
            elapsed = timec - timea;
        } else {
            BayesPm pm = new BayesPm(dag, 3, 3);
            MlBayesIm im = new MlBayesIm(pm, MlBayesIm.RANDOM);
            DataSet data = im.simulateData(numCases, false, tiers);
            System.out.println("Finishing simulation");
            long time2 = System.currentTimeMillis();
            out.println("Elapsed (simulating the data): " + (time2 - time1) + " ms");
            long time3 = System.currentTimeMillis();
            BDeuScore score = new BDeuScore(data);
            score.setStructurePrior(1);
            score.setSamplePrior(1);
            System.out.println(new Date());
            System.out.println("\nStarting FGES");
            long timea = System.currentTimeMillis();
            Fges fges = new Fges(score);
            // fges.setVerbose(false);
            fges.setNumPatternsToStore(0);
            fges.setOut(System.out);
            fges.setFaithfulnessAssumed(faithfulness);
            fges.setCycleBound(-1);
            long timeb = System.currentTimeMillis();
            estPattern = fges.search();
            long timec = System.currentTimeMillis();
            out.println("Time consructing BDeu score " + (timea - time3) + " ms");
            out.println("Time for FGES constructor " + (timeb - timea) + " ms");
            out.println("Time for FGES search " + (timec - timea) + " ms");
            out.println();
            elapsed = timec - timea;
        }
        System.out.println("Done with FGES");
        System.out.println(new Date());
        // System.out.println("Replacing nodes");d
        // 
        // estPattern = GraphUtils.replaceNodes(estPattern, dag.getNodes());
        // System.out.println("Calculating degree");
        // 
        // double degree = GraphUtils.degree(estPattern);
        // degrees.add(degree);
        // 
        // out.println("Degree out output graph = " + degree);
        double[] comparison = new double[4];
        // int adjFn = GraphUtils.countAdjErrors(pattern, estPattern);
        // int adjFp = GraphUtils.countAdjErrors(estPattern, pattern);
        // int trueAdj = pattern.getNumEdges();
        // 
        // comparison[0] = trueAdj / (double) (trueAdj + adjFp);
        // comparison[1] = trueAdj / (double) (trueAdj + adjFn);
        System.out.println("Counting misclassifications.");
        estPattern = GraphUtils.replaceNodes(estPattern, pattern.getNodes());
        int[][] counts = GraphUtils.edgeMisclassificationCounts(pattern, estPattern, false);
        allCounts.add(counts);
        System.out.println(new Date());
        int sumRow = counts[4][0] + counts[4][3] + counts[4][5];
        int sumCol = counts[0][3] + counts[4][3] + counts[5][3] + counts[7][3];
        int trueArrow = counts[4][3];
        int sumTrueAdjacencies = 0;
        for (int i = 0; i < 7; i++) {
            for (int j = 0; j < 5; j++) {
                sumTrueAdjacencies += counts[i][j];
            }
        }
        int falsePositiveAdjacencies = 0;
        for (int j = 0; j < 5; j++) {
            falsePositiveAdjacencies += counts[7][j];
        }
        int falseNegativeAdjacencies = 0;
        for (int i = 0; i < 5; i++) {
            falseNegativeAdjacencies += counts[i][5];
        }
        comparison[0] = sumTrueAdjacencies / (double) (sumTrueAdjacencies + falsePositiveAdjacencies);
        comparison[1] = sumTrueAdjacencies / (double) (sumTrueAdjacencies + falseNegativeAdjacencies);
        comparison[2] = trueArrow / (double) sumCol;
        comparison[3] = trueArrow / (double) sumRow;
        comparisons.add(comparison);
        out.println(GraphUtils.edgeMisclassifications(counts));
        out.println(precisionRecall(comparison));
        elapsedTimes.add(elapsed);
        out.println("\nElapsed: " + elapsed + " ms");
    }
    printAverageConfusion("Average", allCounts);
    printAveragePrecisionRecall(comparisons);
    printAverageStatistics(elapsedTimes, degrees);
    out.close();
}

Example 18 with LargeScaleSimulation

use of edu.cmu.tetrad.sem.LargeScaleSimulation in project tetrad by cmu-phil.

the class TestGeneralBootstrapTest method testFGESc.

@Test
public void testFGESc() {
    int penaltyDiscount = 2;
    boolean faithfulnessAssumed = false;
    int maxDegree = -1;
    int numVars = 20;
    int edgesPerNode = 2;
    int numLatentConfounders = 0;
    int numCases = 50;
    int numBootstrapSamples = 5;
    boolean verbose = true;
    Graph dag = makeContinuousDAG(numVars, numLatentConfounders, edgesPerNode);
    System.out.println("Truth Graph:");
    System.out.println(dag.toString());
    int[] causalOrdering = new int[numVars];
    for (int i = 0; i < numVars; i++) {
        causalOrdering[i] = i;
    }
    LargeScaleSimulation simulator = new LargeScaleSimulation(dag, dag.getNodes(), causalOrdering);
    DataSet data = simulator.simulateDataFisher(numCases);
    Parameters parameters = new Parameters();
    parameters.set("penaltyDiscount", penaltyDiscount);
    parameters.set("faithfulnessAssumed", faithfulnessAssumed);
    parameters.set("maxDegree", maxDegree);
    parameters.set("numPatternsToStore", 0);
    parameters.set("verbose", verbose);
    ScoreWrapper score = new SemBicScore();
    Algorithm algorithm = new Fges(score);
    GeneralBootstrapTest bootstrapTest = new GeneralBootstrapTest(data, algorithm, numBootstrapSamples);
    bootstrapTest.setVerbose(verbose);
    bootstrapTest.setParameters(parameters);
    bootstrapTest.setEdgeEnsemble(BootstrapEdgeEnsemble.Highest);
    Graph resultGraph = bootstrapTest.search();
    System.out.println("Estimated Graph:");
    System.out.println(resultGraph.toString());
    // Adjacency Confusion Matrix
    int[][] adjAr = GeneralBootstrapTest.getAdjConfusionMatrix(dag, resultGraph);
    printAdjConfusionMatrix(adjAr);
    // Edge Type Confusion Matrix
    int[][] edgeAr = GeneralBootstrapTest.getEdgeTypeConfusionMatrix(dag, resultGraph);
    printEdgeTypeConfusionMatrix(edgeAr);
}

Example 19 with LargeScaleSimulation

use of edu.cmu.tetrad.sem.LargeScaleSimulation in project tetrad by cmu-phil.

the class TestGeneralBootstrapTest method testFCIc.

@Test
public void testFCIc() {
    int penaltyDiscount = 2;
    int depth = 3;
    int maxPathLength = -1;
    int numVars = 20;
    int edgesPerNode = 2;
    int numLatentConfounders = 2;
    int numCases = 50;
    int numBootstrapSamples = 5;
    boolean verbose = true;
    Graph dag = makeContinuousDAG(numVars, numLatentConfounders, edgesPerNode);
    DagToPag dagToPag = new DagToPag(dag);
    Graph truePag = dagToPag.convert();
    System.out.println("Truth PAG_of_the_true_DAG Graph:");
    System.out.println(truePag.toString());
    int[] causalOrdering = new int[numVars];
    for (int i = 0; i < numVars; i++) {
        causalOrdering[i] = i;
    }
    LargeScaleSimulation simulator = new LargeScaleSimulation(dag, dag.getNodes(), causalOrdering);
    DataSet data = simulator.simulateDataFisher(numCases);
    Parameters parameters = new Parameters();
    parameters.set("penaltyDiscount", penaltyDiscount);
    parameters.set("depth", depth);
    parameters.set("maxPathLength", maxPathLength);
    parameters.set("numPatternsToStore", 0);
    parameters.set("verbose", verbose);
    IndependenceWrapper test = new FisherZ();
    Fci algorithm = new Fci(test);
    GeneralBootstrapTest bootstrapTest = new GeneralBootstrapTest(data, algorithm, numBootstrapSamples);
    bootstrapTest.setVerbose(verbose);
    bootstrapTest.setParameters(parameters);
    bootstrapTest.setEdgeEnsemble(BootstrapEdgeEnsemble.Preserved);
    // bootstrapTest.setParallelMode(false);
    Graph resultGraph = bootstrapTest.search();
    System.out.println("Estimated PAG_of_the_true_DAG Graph:");
    System.out.println(resultGraph.toString());
    // Adjacency Confusion Matrix
    int[][] adjAr = GeneralBootstrapTest.getAdjConfusionMatrix(truePag, resultGraph);
    printAdjConfusionMatrix(adjAr);
    // Edge Type Confusion Matrix
    int[][] edgeAr = GeneralBootstrapTest.getEdgeTypeConfusionMatrix(truePag, resultGraph);
    printEdgeTypeConfusionMatrix(edgeAr);
}

Example 20 with LargeScaleSimulation

use of edu.cmu.tetrad.sem.LargeScaleSimulation in project tetrad by cmu-phil.

the class TestGeneralBootstrapTest method testGFCIc.

@Test
public void testGFCIc() {
    int penaltyDiscount = 2;
    boolean faithfulnessAssumed = false;
    int maxDegree = -1;
    int numVars = 20;
    int edgesPerNode = 2;
    int numLatentConfounders = 2;
    int numCases = 50;
    int numBootstrapSamples = 5;
    boolean verbose = true;
    Graph dag = makeContinuousDAG(numVars, numLatentConfounders, edgesPerNode);
    DagToPag dagToPag = new DagToPag(dag);
    Graph truePag = dagToPag.convert();
    System.out.println("Truth PAG_of_the_true_DAG Graph:");
    System.out.println(truePag.toString());
    int[] causalOrdering = new int[numVars];
    for (int i = 0; i < numVars; i++) {
        causalOrdering[i] = i;
    }
    LargeScaleSimulation simulator = new LargeScaleSimulation(dag, dag.getNodes(), causalOrdering);
    DataSet data = simulator.simulateDataFisher(numCases);
    Parameters parameters = new Parameters();
    parameters.set("penaltyDiscount", penaltyDiscount);
    parameters.set("faithfulnessAssumed", faithfulnessAssumed);
    parameters.set("maxDegree", maxDegree);
    parameters.set("numPatternsToStore", 0);
    parameters.set("verbose", verbose);
    ScoreWrapper score = new SemBicScore();
    IndependenceWrapper test = new FisherZ();
    Algorithm algorithm = new Gfci(test, score);
    GeneralBootstrapTest bootstrapTest = new GeneralBootstrapTest(data, algorithm, numBootstrapSamples);
    bootstrapTest.setVerbose(verbose);
    bootstrapTest.setParameters(parameters);
    bootstrapTest.setEdgeEnsemble(BootstrapEdgeEnsemble.Highest);
    Graph resultGraph = bootstrapTest.search();
    System.out.println("Estimated PAG_of_the_true_DAG Graph:");
    System.out.println(resultGraph.toString());
    // Adjacency Confusion Matrix
    int[][] adjAr = GeneralBootstrapTest.getAdjConfusionMatrix(truePag, resultGraph);
    printAdjConfusionMatrix(adjAr);
    // Edge Type Confusion Matrix
    int[][] edgeAr = GeneralBootstrapTest.getEdgeTypeConfusionMatrix(truePag, resultGraph);
    printEdgeTypeConfusionMatrix(edgeAr);
}