Examples with TetradVector edu.cmu.tetrad.util.TetradVector used on opensource projects

Example 36 with TetradVector

use of edu.cmu.tetrad.util.TetradVector in project tetrad by cmu-phil.

the class TimeSeriesUtils method ar.

/**
 * @return the VAR residuals of the given time series with the given number
 * of lags. That is, every variable at the model lag is regressed onto every
 * variable at previous lags, up to the given number of lags, and the
 * residuals of these regressions for each variable are returned.
 */
public static DataSet ar(DataSet timeSeries, int numLags) {
    DataSet timeLags = createLagData(timeSeries, numLags);
    List<Node> regressors = new ArrayList<>();
    for (int i = timeSeries.getNumColumns(); i < timeLags.getNumColumns(); i++) {
        regressors.add(timeLags.getVariable(i));
    }
    Regression regression = new RegressionDataset(timeLags);
    // Regression regression = new RegressionDatasetGeneralized(timeLags);
    TetradMatrix residuals = new TetradMatrix(timeLags.getNumRows(), timeSeries.getNumColumns());
    for (int i = 0; i < timeSeries.getNumColumns(); i++) {
        Node target = timeLags.getVariable(i);
        RegressionResult result = regression.regress(target, regressors);
        TetradVector residualsColumn = result.getResiduals();
        // residuals.viewColumn(i).assign(residualsColumn);
        residuals.assignColumn(i, residualsColumn);
    }
    return ColtDataSet.makeContinuousData(timeSeries.getVariables(), residuals);
}

Example 37 with TetradVector

use of edu.cmu.tetrad.util.TetradVector in project tetrad by cmu-phil.

the class TimeSeriesUtils method structuralVar.

public static VarResult structuralVar(DataSet timeSeries, int numLags) {
    DataSet timeLags = TimeSeriesUtils.createLagData(timeSeries, numLags);
    IKnowledge knowledge = timeLags.getKnowledge().copy();
    for (int i = 0; i <= numLags; i++) {
        knowledge.setTierForbiddenWithin(i, true);
    }
    Score score;
    if (timeLags.isDiscrete()) {
        score = new BDeuScore(timeLags);
    } else if (timeLags.isContinuous()) {
        SemBicScore semBicScore = new SemBicScore(new CovarianceMatrixOnTheFly(timeLags));
        semBicScore.setPenaltyDiscount(2.0);
        score = semBicScore;
    } else {
        throw new IllegalArgumentException("Mixed data set");
    }
    Fges search = new Fges(score);
    search.setKnowledge(knowledge);
    Graph graph = search.search();
    // want to collapse graph here...
    Graph collapsedVarGraph = new EdgeListGraph(timeSeries.getVariables());
    for (Edge edge : graph.getEdges()) {
        String node1_before = edge.getNode1().getName();
        String node2_before = edge.getNode2().getName();
        String node1_after = node1_before.substring(0, node1_before.indexOf("."));
        String node2_after = node2_before.substring(0, node2_before.indexOf("."));
        Node node1 = collapsedVarGraph.getNode(node1_after);
        Node node2 = collapsedVarGraph.getNode(node2_after);
        Edge _edge = new Edge(node1, node2, edge.getEndpoint1(), edge.getEndpoint2());
        if (!collapsedVarGraph.containsEdge(_edge)) {
            collapsedVarGraph.addEdge(_edge);
        }
    }
    TetradMatrix residuals = new TetradMatrix(timeLags.getNumRows(), timeSeries.getNumColumns());
    Regression regression = new RegressionDataset(timeLags);
    for (int i = 0; i < timeSeries.getNumColumns(); i++) {
        Node target = timeLags.getVariable(i);
        List<Node> regressors = new ArrayList<>();
        // timelags.getVariable(i).
        for (int j = 0; j <= 0; /*numLags*/
        j++) {
            Node variable = timeLags.getVariable(i + j * timeSeries.getNumColumns());
            regressors.addAll(graph.getParents(variable));
        }
        RegressionResult result = regression.regress(target, regressors);
        TetradVector residualsColumn = result.getResiduals();
        // residuals.viewColumn(i).assign(residualsColumn);
        residuals.assignColumn(i, residualsColumn);
    }
    return new VarResult(ColtDataSet.makeContinuousData(timeSeries.getVariables(), residuals), collapsedVarGraph);
}

Example 38 with TetradVector

use of edu.cmu.tetrad.util.TetradVector in project tetrad by cmu-phil.

the class SemUpdater method getUpdatedSemIm.

/**
 * See http://en.wikipedia.org/wiki/Multivariate_normal_distribution.
 */
public SemIm getUpdatedSemIm() {
    Algebra algebra = new Algebra();
    // First manipulate the old semIm.
    SemIm manipulatedSemIm = getManipulatedSemIm();
    // Get out the means and implied covariances.
    double[] means = new double[manipulatedSemIm.getVariableNodes().size()];
    for (int i = 0; i < means.length; i++) {
        means[i] = manipulatedSemIm.getMean(manipulatedSemIm.getVariableNodes().get(i));
    }
    DoubleMatrix1D mu = new DenseDoubleMatrix1D(means);
    DoubleMatrix2D sigma = new DenseDoubleMatrix2D(manipulatedSemIm.getImplCovar(true).toArray());
    // Updating on x2 = a.
    SemEvidence evidence = getEvidence();
    List nodesInEvidence = evidence.getNodesInEvidence();
    int[] x2 = new int[nodesInEvidence.size()];
    DoubleMatrix1D a = new DenseDoubleMatrix1D(nodesInEvidence.size());
    for (int i = 0; i < nodesInEvidence.size(); i++) {
        Node _node = (Node) nodesInEvidence.get(i);
        x2[i] = evidence.getNodeIndex(_node);
    }
    for (int i = 0; i < nodesInEvidence.size(); i++) {
        int j = evidence.getNodeIndex((Node) nodesInEvidence.get(i));
        a.set(i, evidence.getProposition().getValue(j));
    }
    // x1 is all the variables.
    // int[] x1 = new int[sigma.rows() - x2.length];
    int[] x1 = new int[sigma.rows()];
    for (int i = 0; i < sigma.rows(); i++) {
        x1[i] = i;
    }
    // int index = -1;
    // for (int i = 0; i < sigma.rows(); i++) {
    // if (Arrays.binarySearch(x2, i) == -1) {
    // x1[++index] = i;
    // }
    // }
    // Calculate sigmaBar. (Don't know how to use it yet.)
    // DoubleMatrix2D sigma11 = sigma.viewSelection(x1, x1);
    DoubleMatrix2D sigma12 = sigma.viewSelection(x1, x2);
    DoubleMatrix2D sigma22 = sigma.viewSelection(x2, x2);
    // DoubleMatrix2D sigma21 = sigma.viewSelection(x2, x1);
    DoubleMatrix2D inv_sigma22 = algebra.inverse(sigma22);
    DoubleMatrix2D temp1 = algebra.mult(sigma12, inv_sigma22);
    // DoubleMatrix2D temp2 = algebra.times(temp1, sigma21.copy());
    // DoubleMatrix2D sigmaBar = sigma11.copy().assign(temp2, Functions.minus);
    // Calculate muBar.
    DoubleMatrix1D mu1 = mu.viewSelection(x1);
    DoubleMatrix1D mu2 = mu.viewSelection(x2);
    DoubleMatrix1D temp4 = a.copy().assign(mu2, Functions.minus);
    DoubleMatrix1D temp5 = algebra.mult(temp1, temp4);
    DoubleMatrix1D muBar = mu1.copy().assign(temp5, Functions.plus);
    // Estimate a SEM with this sigmaBar and muBar.
    // List variableNodes = manipulatedSemIm.getVariableNodes();
    // String[] varNames = new String[variableNodes.size()];
    // 
    // for (int i = 0; i < variableNodes.size(); i++) {
    // varNames[i] = ((Node) variableNodes.get(i)).getNode();
    // }
    // System.out.println(sigmaBar);
    // 
    // CovarianceMatrix covMatrix = new CovarianceMatrix(varNames,
    // sigmaBar, 100);
    // SemPm semPm = manipulatedSemIm.getEstIm();
    // SemEstimator estimator = new SemEstimator(covMatrix, semPm);
    // estimator.estimate();
    // SemIm semIm = estimator.getEstimatedSem();
    // semIm.setMeanValues(muBar.toArray());
    // return semIm;
    DoubleMatrix2D sigma2 = new DenseDoubleMatrix2D(manipulatedSemIm.getErrCovar().toArray());
    // }
    return manipulatedSemIm.updatedIm(new TetradMatrix(sigma2.toArray()), new TetradVector(muBar.toArray()));
}

Example 39 with TetradVector

use of edu.cmu.tetrad.util.TetradVector in project tetrad by cmu-phil.

the class SemOptimizerEm method maximization.

private void maximization() {
    List<Node> nodes = graph.getNodes();
    for (Node node : graph.getNodes()) {
        if (node.getNodeType() == NodeType.ERROR) {
            continue;
        }
        int idx = nodes.indexOf(node);
        double variance = expectedCov.get(idx, idx);
        if (parents[idx] != null) {
            for (int i = 0; i < parents[idx].length; i++) {
                int idx2 = parents[idx][i];
                nodeParentsCov[idx][i] = expectedCov.get(idx, idx2);
                for (int j = i; j < parents[idx].length; j++) {
                    int idx3 = parents[idx][j];
                    parentsCov[idx][i][j] = expectedCov.get(idx2, idx3);
                    parentsCov[idx][j][i] = expectedCov.get(idx3, idx2);
                }
            }
            TetradVector coefs = new TetradMatrix(parentsCov[idx]).inverse().times(new TetradVector(nodeParentsCov[idx]));
            for (int i = 0; i < coefs.size(); i++) {
                if (semIm.getSemPm().getParameter(nodes.get(parents[idx][i]), node) != null && !semIm.getSemPm().getParameter(nodes.get(parents[idx][i]), node).isFixed()) {
                    semIm.setEdgeCoef(nodes.get(parents[idx][i]), node, coefs.get(i));
                }
            }
            variance -= new TetradVector(nodeParentsCov[idx]).dotProduct(coefs);
        }
        if (!semIm.getSemPm().getParameter(errorParent[idx], errorParent[idx]).isFixed()) {
            semIm.setErrCovar(errorParent[idx], variance);
        }
    }
}

Example 40 with TetradVector

use of edu.cmu.tetrad.util.TetradVector in project tetrad by cmu-phil.

the class SemOptimizerRegression method optimize.

// ============================PUBLIC METHODS==========================//
/**
 * Fit the freeParameters by doing local regressions.
 */
public void optimize(SemIm semIm) {
    if (numRestarts != 1) {
        throw new IllegalArgumentException("Number of restarts must be 1 for this method.");
    }
    TetradMatrix covar = semIm.getSampleCovar();
    if (covar == null) {
        throw new NullPointerException("Sample covar has not been set.");
    }
    SemGraph graph = semIm.getSemPm().getGraph();
    List<Node> nodes = graph.getNodes();
    for (Node node : nodes) {
        if (node.getNodeType() != NodeType.MEASURED) {
            continue;
        }
        if (!graph.isParameterizable(node))
            continue;
        int idx = nodes.indexOf(node);
        List<Node> parents = graph.getParents(node);
        for (int i = 0; i < parents.size(); i++) {
            Node nextParent = parents.get(i);
            if (nextParent.getNodeType() == NodeType.ERROR) {
                // errorParent = nextParent;
                parents.remove(nextParent);
                break;
            }
        }
        double variance = covar.get(idx, idx);
        if (parents.size() > 0) {
            TetradVector nodeParentsCov = new TetradVector(parents.size());
            TetradMatrix parentsCov = new TetradMatrix(parents.size(), parents.size());
            for (int i = 0; i < parents.size(); i++) {
                int idx2 = nodes.indexOf(parents.get(i));
                nodeParentsCov.set(i, covar.get(idx, idx2));
                for (int j = i; j < parents.size(); j++) {
                    int idx3 = nodes.indexOf(parents.get(j));
                    parentsCov.set(i, j, covar.get(idx2, idx3));
                    parentsCov.set(j, i, covar.get(idx2, idx3));
                }
            }
            TetradVector b = parentsCov.inverse().times(nodeParentsCov);
            variance -= nodeParentsCov.dotProduct(b);
            for (int i = 0; i < b.size(); i++) {
                int idx2 = nodes.indexOf(parents.get(i));
                semIm.setParamValue(nodes.get(idx2), node, b.get(i));
            }
        }
        semIm.setParamValue(node, node, variance);
    }
    TetradLogger.getInstance().log("optimization", "FML = " + semIm.getScore());
}