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

Example 6 with RandomUtil

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

the class TestBoxDataSet method testDiscrete.

@Test
public void testDiscrete() {
    int rows = 10;
    int cols = 5;
    List<Node> variables = new LinkedList<>();
    for (int i = 0; i < cols; i++) {
        DiscreteVariable variable = new DiscreteVariable("X" + (i + 1), 3);
        variables.add(variable);
    }
    DataSet dataSet = new BoxDataSet(new DoubleDataBox(rows, variables.size()), variables);
    RandomUtil randomUtil = RandomUtil.getInstance();
    for (int i = 0; i < rows; i++) {
        for (int j = 0; j < cols; j++) {
            dataSet.setInt(i, j, randomUtil.nextInt(3));
        }
    }
    BoxDataSet _dataSet = new BoxDataSet((BoxDataSet) dataSet);
    assertEquals(dataSet, _dataSet);
}

Example 7 with RandomUtil

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

the class Hsim method hybridsimulate.

// **************Public methods***********************//
public DataSet hybridsimulate() {
    /**
     *this needs to be made general, rather than only for two specific names nodes*
     */
    if (verbose)
        System.out.println("Finding a Markov blanket for resimulated nodes");
    // initialize an empty set of nodes;
    Set<Node> mbAll = new HashSet<Node>();
    // init set for adding
    Set<Node> mbAdd = new HashSet<Node>();
    for (Node node : simnodes) {
        // find mb for that node
        mbAdd = mb(mydag, node);
        // use .addAll to add this mb to the set
        mbAll.addAll(mbAdd);
    }
    // make sure all the simnodes are in mbAll! a disconnected node could cause errors later otherwise
    mbAll.addAll(simnodes);
    if (verbose)
        System.out.println("The Markov Blanket is " + mbAll);
    /**
     *Find the subgraph for the resimulated variables and their markov blanket*
     */
    if (verbose)
        System.out.println("Finding a subgraph over the Markov Blanket and Resimulated Nodes");
    // need a List as input for subgraph method, but mbAll is a Set
    List<Node> mbListAll = new ArrayList<Node>(mbAll);
    Graph subgraph = mydag.subgraph(mbListAll);
    /**
     *Learn an instantiated model over the subgraph*
     */
    if (verbose)
        System.out.println("Learning an instantiated model for the subgraph");
    // learn a dirichlet IM for the subgraph using dataSet
    BayesPm subgraphPM = new BayesPm(subgraph);
    DirichletBayesIm subgraphIM = DirichletBayesIm.symmetricDirichletIm(subgraphPM, 1.0);
    DirichletEstimator estimator = new DirichletEstimator();
    DirichletBayesIm fittedsubgraphIM = estimator.estimate(subgraphIM, data);
    /**
     *Use the learned instantiated subgraph model to create the resimulated data*
     */
    if (verbose)
        System.out.println("Starting resimulation loop");
    // loop through each row of the data set, conditioning and drawing values each time.
    for (int row = 0; row < data.getNumRows(); row++) {
        // create a new evidence object
        Evidence evidence = Evidence.tautology(fittedsubgraphIM);
        // need to define the set of variables being conditioned upon. Start with the outer set of MB
        Set<Node> mbOuter = mbAll;
        // need to remove the whole set of starters, not just some X and Y... how do? loop a .remove?
        for (Node node : simnodes) {
            mbOuter.remove(node);
        }
        // THIS SHOULD ALL BE INSIDE ANOTHER LOOP THROUGH THE RESIM VARS:
        // this actually needs to be more careful than a for each. I think a causal ordering of resim should be used?
        Set<Node> conditionNodes = mbOuter;
        for (Node node : simnodes) {
            // loop through all the nodes being conditioned upon, and set their values in the evidence prop
            for (Node i : conditionNodes) {
                int nodeIndex = evidence.getNodeIndex(i.getName());
                // how do i get the category index from a value in the data?
                // int catIndex =
                int nodeColumn = data.getColumn(i);
                // Pray to whoever you can think of that the CategoryIndex is just the int in the data
                // According to this comment in the DataSet class, for the getInt method, we can do this:
                // "For discrete variables, this returns the category index of the datum for the variable at that column."
                evidence.getProposition().setCategory(nodeIndex, data.getInt(row, nodeColumn));
            }
            // use the new Evidence object to create the updater
            RowSummingExactUpdater conditionUpdate = new RowSummingExactUpdater(fittedsubgraphIM, evidence);
            // NEED THIS TO WORK FOR MORE THAN NODEX, needs to be arbitrary and looping
            // use the updater to create the marginal distribution for nodeX:
            // need to get nodeX's int index first
            int nodeIndex = evidence.getNodeIndex(node.getName());
            // ===complain if no node of that name is found, which makes nodeIndex = -1
            if (nodeIndex == -1) {
                throw new IllegalArgumentException("Variable " + node.getName() + " was not found.");
            }
            // ===for bug checking====
            // if (verbose) System.out.println(node.getName());
            // if (verbose) System.out.println(nodeIndex);
            // then need to identify all of nodeX's categories so we can iterate through them
            // I can't figure out a nice way to identify this generally, so we're going to cross out fingers
            // and hope that it's just 0, 1, 2, 3... n-1, with n-1 being the largest int category
            // that would correspond to n different categories
            // so, we're gonna see how many categories there are, call that n, and iterate from 0 to n-1
            int numCat = evidence.getNumCategories(nodeIndex);
            // we generate a random number between 0 and 1, and then count prob mass for each cat until we hit it
            RandomUtil random = RandomUtil.getInstance();
            double cutoff = random.nextDouble();
            // if (verbose) System.out.println(cutoff);
            // ****** turns out, this needs to be generalized outside of just X and Y as well. doh!*******//
            // for (resimvars) {do the next stuff} //how to iterate through them? order matters, need causal ordering
            double sum = 0.0;
            // initialize the int for the new value of nodeX
            int newValue = -99;
            // now iterate through the categories to see which one owns that probability mass real estate
            for (int i = 0; i < numCat; i++) {
                // for each category, calc the marginal conditional prob of nodeX having that value
                double probability = conditionUpdate.getMarginal(nodeIndex, i);
                // if (verbose) System.out.println("cat " + i + " prob " + probability);
                sum += probability;
                if (sum >= cutoff) {
                    newValue = i;
                    break;
                }
            }
            // then set the value of nodeX to newXvalue for this row
            // if (verbose) System.out.println(data.getInt(row,data.getColumn(nodeX)) + " old vs new " + newXvalue);
            data.setInt(row, data.getColumn(node), newValue);
            // if (verbose) System.out.println(" and again?: " + data.getInt(row,data.getColumn(nodeX)) + " old vs new " + newXvalue);
            // at the end, at this node to the conditioning set
            conditionNodes.add(node);
        }
    }
    return data;
}

Example 8 with RandomUtil

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

the class MlBayesImObs method simulateData.

/**
 * Simulates a sample with the given sample size.
 *
 * @param sampleSize      the sample size.
 * @param seed            the random number generator seed allows you
 *                        recreate the simulated data by passing in the same
 *                        seed (so you don't have to store the sample data
 * @return the simulated sample as a DataSet.
 */
public DataSet simulateData(int sampleSize, long seed, boolean latentDataSaved) {
    RandomUtil random = RandomUtil.getInstance();
    long _seed = random.getSeed();
    random.setSeed(seed);
    DataSet dataSet = simulateData(sampleSize, latentDataSaved);
    random.revertSeed(_seed);
    return dataSet;
}

Example 9 with RandomUtil

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

the class BayesPm method pickNumVals.

private static int pickNumVals(int lowerBound, int upperBound) {
    if (lowerBound < 2) {
        throw new IllegalArgumentException("Lower bound must be >= 2: " + lowerBound);
    }
    if (upperBound < lowerBound) {
        throw new IllegalArgumentException("Upper bound for number of categories must be >= lower " + "bound.");
    }
    int difference = upperBound - lowerBound;
    RandomUtil randomUtil = RandomUtil.getInstance();
    return randomUtil.nextInt(difference + 1) + lowerBound;
}

Example 10 with RandomUtil

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

the class DirichletBayesIm method simulateData.

/**
 * Simulates a random sample with the number of cases equal to
 * <code>sampleSize</code>.
 *
 * @param sampleSize      the sample size.
 * @param seed            the random number generator seed allows you
 *                        recreate the simulated data by passing in the same
 *                        seed (so you don't have to store the sample data
 * @param latentDataSaved true iff data for latent variables should be
 *                        included in the simulated data set.
 * @return the simulated sample as a DataSet.
 */
public DataSet simulateData(int sampleSize, long seed, boolean latentDataSaved) {
    RandomUtil random = RandomUtil.getInstance();
    long _seed = random.getSeed();
    random.setSeed(seed);
    DataSet dataSet = simulateData(sampleSize, latentDataSaved);
    random.revertSeed(_seed);
    return dataSet;
}