Examples with GLMParameters hex.glm.GLMModel.GLMParameters used on opensource projects

Example 36 with GLMParameters

use of hex.glm.GLMModel.GLMParameters in project h2o-3 by h2oai.

the class GLMBasicTestRegression method testSparse.

@Test
public void testSparse() {
    double[] exp_coefs = new double[] { 0.0233151691783671, -0.00543776852277619, -0.0137359312181047, 0.00770037200907652, 0.0328856331139761, -0.0242845468071283, -0.0101698117745265, 0.00868844870137727, 0.000349121513384513, -0.0106962199761512, -0.00705001448025939, 0.00821574637914086, 0.00601015905212279, 0.0021278467162546, -0.0233079168835112, 0.00535473896013676, -0.00897667301004576, 0.00788272228017582, 0.00237442711371947, -0.013136425134371, 0.00134003869245749, 0.0240118046676911, 0.000607214787933269, -0.0112908513868027, 0.000443119443631777, 0.00749330452744921, -0.00558704122833295, 0.000533036850835694, 0.0130008059852934, -4.40634889376063e-05, -0.00580285872202347, 0.0117029111583238, -0.00685480666428133, 0.00809526311326634, -0.0088567165389072, -0.0363126456378731, -0.00267237519808936, -0.01669554043682, 0.00556943053195684, 0.0178196407614288, -0.000903204442155076, -0.0085363297586185, -0.00421147221966977, -0.00828702756129772, 0.017027928644479, 0.00710126315700672, 0.019819043342772, -0.0165232485929677, 0.00439570108491533, 0.0188325734374437, 0.00799712968759025, -0.0100388875424171, -0.0062415137856855, -0.00258013659839137, -6.58516379178382e-05, 0.0135032332096949, -0.00776869619293087, -0.00544035128543343, -0.0110626226606883, -0.00768490011210769, -0.00684181016695251, -0.0144627862333649, -0.0262830557415184, -0.0102290180164706, 0.00368252955770187, 0.015824495748353, 0.00383484095683782, 0.0151193905626625, -0.00615077094420626, 0.0142842231522414, 0.00150448184871646, 0.0521491615912011, 0.0128661232226479, 0.00225580439739044, -0.0117476427864401, -0.0059792656068627, 0.000787012740598272, 0.00255419488737936, 0.00406033118385186, 0.0102551045653601, 0.00423949002681056, -0.0116986428989079, 0.00232448128787425, -0.00296198808290357, -0.00793738689381332, -0.000771158906679964, 0.00435708760153937, -0.0138922325725763, 0.00264561130131037, -0.0156128295187466, -0.0102023187068811, 0.0074744189329328, 0.0102377749189598, -0.0304739969497646, 0.00692556661464647, 0.00151065993974025, 0.0133704258946895, -0.0167391228441308, 0.0111804482435337, -0.0062469732087272, -0.00930165243463748, -0.00418698783410104, 0.00190918091726462, 0.00632982717347925, -0.00277608255480933, -0.00175463261672652, -0.00267223587651978, -0.00329264073314718, 0.000960091877616874, -0.00946014799557438, -0.0112302467393988, -0.00870512647578646, -0.00238582834931644, -0.0100845163232815, -0.00675861103174491, -0.000689229731411459, 0.0127651588318169, -0.0062753105816655, -0.00240575758827749, 0.00439570108491531, 0.00934971690544427, -0.0184380964678117, -0.00474253892124699, 0.00522916014066936, -0.0105148336464531, 0.0088372219244051, 0.0100429095740915, -0.0107657032259033, -0.00512476269437683, -0.00558487620671732, -0.000637298812579742, -0.00118460090105795, -0.00369801350318738, -0.00556276860695209, 0.00789011470305446, -0.00248367841256358, 0.00677762904717052, -0.00640135771848287, 0.00797532960057465, -0.00117508910987595, 0.000986931150778778, -0.0148237721063735, 0.0053001635341953, -0.0139698571439444, -0.0172255105183439, -0.0177416268392445, -0.0107062660197562, -0.00735448768491512, -0.00418482390542493, 0.00933957546887131, -0.00761657876743367, 0.0107862806984669, 6.99667442150322e-05, -0.00151054027221715, 0.00941377216029456, 0.0112882845381545, 0.0014423575345095, 0.00845773223444363, -0.00675939077916714, -0.00329806028742896, 0.000276998824889068, 0.00206337643122044, -0.00173085772672239, 0.00169616445468346, 0.00281297187309321, -0.0152343998246272, 0.0126261762792184, -0.000224959505615703, -0.00476466349783071, -0.0102541605421868, -0.000561674281900828, 0.00367777757696579, -0.000960272764476094, 0.00255704179717728, -0.000696266184051808, 0.0470920125432207, 0.0115016691642458, -0.00287666464467251, -0.00132912286075637, 0.00201932482935891, 0.00119899092739739, 0.00380417340899902, -0.00394363983208331, -0.00294543812868618, -1.77894150438862e-05, -0.00455002740798846, 0.000613307426862812, 0.00348274063618593, 0.00161877234851832, 0.0231608701706833, -0.00390062462708628, 0.00244047437999614, -0.00143984617445982, -0.00221831741496412, -0.00744853810342609, -0.00575689075773469, -0.00567890661011033, 0.00384589889309526, -0.00173241442296732, -0.00526995531653655, -0.00310819786514896, 0.00740596461822877, -0.0790037392468225, 0.0239744234187787, 0.0514310481067108, 0.034335426530007, 0.0254604884688754, 0.0531375235023675, -0.0228335779154641, 0.546865402727144 };
    GLMModel model1 = null;
    GLMParameters parms = new GLMParameters(Family.gaussian);
    _airlinesMM.add("weights", _airlinesMM.anyVec().makeCon(1.0));
    DKV.put(_airlinesMM._key, _airlinesMM);
    parms._weights_column = "weights";
    parms._train = _airlinesMM._key;
    parms._lambda = new double[] { 1e-2 };
    parms._alpha = new double[] { 0 };
    parms._solver = Solver.IRLSM;
    parms._ignored_columns = new String[] { "C1" };
    //    parms._remove_collinear_columns = true;
    parms._response_column = "IsDepDelayed";
    parms._standardize = true;
    parms._objective_epsilon = 0;
    parms._gradient_epsilon = 1e-10;
    parms._max_iterations = 1000;
    parms._missing_values_handling = DeepLearningModel.DeepLearningParameters.MissingValuesHandling.Skip;
    try {
        model1 = new GLM(parms).trainModel().get();
        for (int i = 0; i < model1._output._coefficient_names.length; ++i) assertEquals(exp_coefs[i], model1._output.getNormBeta()[i], Math.abs(exp_coefs[i]) * 1e-8);
    } finally {
        if (model1 != null)
            model1.delete();
    }
}

Example 37 with GLMParameters

use of hex.glm.GLMModel.GLMParameters in project h2o-3 by h2oai.

the class GLMBasicTestRegression method testPValuesTweedie.

@Test
public void testPValuesTweedie() {
    //    Call:
    //    glm(formula = Infections ~ ., family = tweedie(var.power = 1.5),
    //      data = D)
    //
    //    Deviance Residuals:
    //    Min       1Q   Median       3Q      Max
    //    -2.6355  -2.0931  -1.8183   0.5046   4.9458
    //
    //    Coefficients:
    //    Estimate Std. Error t value Pr(>|t|)
    //    (Intercept)       1.05665    0.11120   9.502  < 2e-16 ***
    //    SwimmerOccas     -0.25891    0.08455  -3.062  0.00241 **
    //    LocationNonBeach -0.22185    0.08393  -2.643  0.00867 **
    //    Age20-24          0.15325    0.10041   1.526  0.12808
    //    Age25-29          0.07624    0.10099   0.755  0.45096
    //    SexMale           0.03908    0.08619   0.453  0.65058
    //      ---
    //      Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
    //
    //    (Dispersion parameter for Tweedie family taken to be 2.896306)
    //
    //    Null deviance: 967.05  on 286  degrees of freedom
    //    Residual deviance: 908.86  on 281  degrees of freedom
    //    AIC: NA
    //
    //    Number of Fisher Scoring iterations: 7
    double[] sderr_exp = new double[] { 0.11120211, 0.08454967, 0.08393315, 0.10041150, 0.10099231, 0.08618960 };
    double[] zvals_exp = new double[] { 9.5021062, -3.0622693, -2.6431794, 1.5262357, 0.7548661, 0.4534433 };
    double[] pvals_exp = new double[] { 9.508400e-19, 2.409514e-03, 8.674149e-03, 1.280759e-01, 4.509615e-01, 6.505795e-01 };
    GLMParameters parms = new GLMParameters(Family.tweedie);
    parms._tweedie_variance_power = 1.5;
    parms._tweedie_link_power = 1 - parms._tweedie_variance_power;
    parms._train = _earinf._key;
    parms._standardize = false;
    parms._lambda = new double[] { 0 };
    parms._alpha = new double[] { 0 };
    parms._response_column = "Infections";
    parms._compute_p_values = true;
    parms._objective_epsilon = 0;
    parms._missing_values_handling = DeepLearningModel.DeepLearningParameters.MissingValuesHandling.Skip;
    GLMModel model = null;
    Frame predict = null;
    try {
        model = new GLM(parms).trainModel().get();
        String[] names_expected = new String[] { "Intercept", "Swimmer.Occas", "Location.NonBeach", "Age.20-24", "Age.25-29", "Sex.Male" };
        String[] names_actual = model._output.coefficientNames();
        HashMap<String, Integer> coefMap = new HashMap<>();
        for (int i = 0; i < names_expected.length; ++i) coefMap.put(names_expected[i], i);
        double[] stder_actual = model._output.stdErr();
        double[] zvals_actual = model._output.zValues();
        double[] pvals_actual = model._output.pValues();
        for (int i = 0; i < sderr_exp.length; ++i) {
            int id = coefMap.get(names_actual[i]);
            assertEquals(sderr_exp[id], stder_actual[i], sderr_exp[id] * 1e-3);
            assertEquals(zvals_exp[id], zvals_actual[i], Math.abs(zvals_exp[id]) * 1e-3);
            assertEquals(pvals_exp[id], pvals_actual[i], Math.max(1e-8, pvals_exp[id]) * 5e-3);
        }
        predict = model.score(parms._train.get());
        Vec.Reader r = predict.vec("StdErr").new Reader();
        for (int i = 0; i < 10; i++) System.out.println(tweedie_se_fit[i] + " ?=? " + r.at(i));
        for (int i = 0; i < tweedie_se_fit.length; ++i) assertEquals(tweedie_se_fit[i], r.at(i), 1e-4);
    } finally {
        if (model != null)
            model.delete();
        if (predict != null)
            predict.delete();
    }
}

Example 38 with GLMParameters

use of hex.glm.GLMModel.GLMParameters in project h2o-3 by h2oai.

the class GLMTest method testProstate.

/**
   * Simple test for binomial family (no regularization, test both lsm solvers).
   * Runs the classical prostate, using dataset with race replaced by categoricals (probably as it's supposed to be?), in any case,
   * it gets to test correct processing of categoricals.
   *
   * Compare against the results from standard R glm implementation.
   * @throws ExecutionException
   * @throws InterruptedException
   */
@Test
public void testProstate() throws InterruptedException, ExecutionException {
    GLMModel model = null, model2 = null, model3 = null, model4 = null;
    Frame fr = parse_test_file("smalldata/glm_test/prostate_cat_replaced.csv");
    try {
        Scope.enter();
        // R results
        //      Coefficients:
        //        (Intercept)           ID          AGE       RACER2       RACER3        DPROS        DCAPS          PSA          VOL      GLEASON
        //          -8.894088     0.001588    -0.009589     0.231777    -0.459937     0.556231     0.556395     0.027854    -0.011355     1.010179
        String[] cfs1 = new String[] { "Intercept", "AGE", "RACE.R2", "RACE.R3", "DPROS", "DCAPS", "PSA", "VOL", "GLEASON" };
        double[] vals = new double[] { -8.14867, -0.01368, 0.32337, -0.38028, 0.55964, 0.49548, 0.02794, -0.01104, 0.97704 };
        GLMParameters params = new GLMParameters(Family.binomial);
        params._response_column = "CAPSULE";
        params._ignored_columns = new String[] { "ID" };
        params._train = fr._key;
        params._lambda = new double[] { 0 };
        params._standardize = false;
        //      params._missing_values_handling = MissingValuesHandling.Skip;
        GLM glm = new GLM(params);
        model = glm.trainModel().get();
        assertTrue(model._output.bestSubmodel().iteration == 5);
        model.delete();
        params._max_iterations = 4;
        glm = new GLM(params);
        model = glm.trainModel().get();
        assertTrue(model._output.bestSubmodel().iteration == 4);
        System.out.println(model._output._model_summary);
        HashMap<String, Double> coefs = model.coefficients();
        System.out.println(coefs);
        for (int i = 0; i < cfs1.length; ++i) assertEquals(vals[i], coefs.get(cfs1[i]), 1e-4);
        assertEquals(512.3, nullDeviance(model), 1e-1);
        assertEquals(378.3, residualDeviance(model), 1e-1);
        assertEquals(371, resDOF(model), 0);
        assertEquals(396.3, aic(model), 1e-1);
        testScoring(model, fr);
        // test scoring
        model.score(fr).delete();
        hex.ModelMetricsBinomial mm = hex.ModelMetricsBinomial.getFromDKV(model, fr);
        hex.AUC2 adata = mm._auc;
        assertEquals(model._output._training_metrics.auc_obj()._auc, adata._auc, 1e-8);
        assertEquals(0.7588625640559653, adata.pr_auc(), 1e-8);
        assertEquals(model._output._training_metrics._MSE, mm._MSE, 1e-8);
        assertEquals(((ModelMetricsBinomialGLM) model._output._training_metrics)._resDev, ((ModelMetricsBinomialGLM) mm)._resDev, 1e-8);
        model.score(fr).delete();
        mm = hex.ModelMetricsBinomial.getFromDKV(model, fr);
        assertEquals(model._output._training_metrics.auc_obj()._auc, adata._auc, 1e-8);
        assertEquals(model._output._training_metrics._MSE, mm._MSE, 1e-8);
        assertEquals(((ModelMetricsBinomialGLM) model._output._training_metrics)._resDev, ((ModelMetricsBinomialGLM) mm)._resDev, 1e-8);
        double prior = 1e-5;
        params._prior = prior;
        // test the same data and model with prior, should get the same model except for the intercept
        glm = new GLM(params);
        model2 = glm.trainModel().get();
        for (int i = 0; i < model2.beta().length - 1; ++i) assertEquals(model.beta()[i], model2.beta()[i], 1e-8);
        assertEquals(model.beta()[model.beta().length - 1] - Math.log(model._ymu[0] * (1 - prior) / (prior * (1 - model._ymu[0]))), model2.beta()[model.beta().length - 1], 1e-10);
        // run with lambda search, check the final submodel
        params._lambda_search = true;
        params._lambda = null;
        params._alpha = new double[] { 0 };
        params._prior = -1;
        params._obj_reg = -1;
        params._max_iterations = 500;
        params._objective_epsilon = 1e-6;
        // test the same data and model with prior, should get the same model except for the intercept
        glm = new GLM(params);
        model3 = glm.trainModel().get();
        double lambda = model3._output._submodels[model3._output._best_lambda_idx].lambda_value;
        params._lambda_search = false;
        params._lambda = new double[] { lambda };
        ModelMetrics mm3 = ModelMetrics.getFromDKV(model3, fr);
        assertEquals("mse don't match, " + model3._output._training_metrics._MSE + " != " + mm3._MSE, model3._output._training_metrics._MSE, mm3._MSE, 1e-8);
        assertEquals("res-devs don't match, " + ((ModelMetricsBinomialGLM) model3._output._training_metrics)._resDev + " != " + ((ModelMetricsBinomialGLM) mm3)._resDev, ((ModelMetricsBinomialGLM) model3._output._training_metrics)._resDev, ((ModelMetricsBinomialGLM) mm3)._resDev, 1e-4);
        fr.add("CAPSULE", fr.remove("CAPSULE"));
        fr.remove("ID").remove();
        DKV.put(fr._key, fr);
        DataInfo dinfo = new DataInfo(fr, null, 1, true, TransformType.NONE, DataInfo.TransformType.NONE, true, false, false, false, false, false);
        model3.score(fr).delete();
        mm3 = ModelMetrics.getFromDKV(model3, fr);
        assertEquals("mse don't match, " + model3._output._training_metrics._MSE + " != " + mm3._MSE, model3._output._training_metrics._MSE, mm3._MSE, 1e-8);
        assertEquals("res-devs don't match, " + ((ModelMetricsBinomialGLM) model3._output._training_metrics)._resDev + " != " + ((ModelMetricsBinomialGLM) mm3)._resDev, ((ModelMetricsBinomialGLM) model3._output._training_metrics)._resDev, ((ModelMetricsBinomialGLM) mm3)._resDev, 1e-4);
        // test the same data and model with prior, should get the same model except for the intercept
        glm = new GLM(params);
        model4 = glm.trainModel().get();
        assertEquals("mse don't match, " + model3._output._training_metrics._MSE + " != " + model4._output._training_metrics._MSE, model3._output._training_metrics._MSE, model4._output._training_metrics._MSE, 1e-6);
        assertEquals("res-devs don't match, " + ((ModelMetricsBinomialGLM) model3._output._training_metrics)._resDev + " != " + ((ModelMetricsBinomialGLM) model4._output._training_metrics)._resDev, ((ModelMetricsBinomialGLM) model3._output._training_metrics)._resDev, ((ModelMetricsBinomialGLM) model4._output._training_metrics)._resDev, 1e-4);
        model4.score(fr).delete();
        ModelMetrics mm4 = ModelMetrics.getFromDKV(model4, fr);
        assertEquals("mse don't match, " + mm3._MSE + " != " + mm4._MSE, mm3._MSE, mm4._MSE, 1e-6);
        assertEquals("res-devs don't match, " + ((ModelMetricsBinomialGLM) mm3)._resDev + " != " + ((ModelMetricsBinomialGLM) mm4)._resDev, ((ModelMetricsBinomialGLM) mm3)._resDev, ((ModelMetricsBinomialGLM) mm4)._resDev, 1e-4);
    //      GLMValidation val2 = new GLMValidationTsk(params,model._ymu,rank(model.beta())).doAll(new Vec[]{fr.vec("CAPSULE"),score.vec("1")})._val;
    //      assertEquals(val.residualDeviance(),val2.residualDeviance(),1e-6);
    //      assertEquals(val.nullDeviance(),val2.nullDeviance(),1e-6);
    } finally {
        fr.delete();
        if (model != null)
            model.delete();
        if (model2 != null)
            model2.delete();
        if (model3 != null)
            model3.delete();
        if (model4 != null)
            model4.delete();
        Scope.exit();
    }
}

Example 39 with GLMParameters

use of hex.glm.GLMModel.GLMParameters in project h2o-3 by h2oai.

the class GLMTest method testAllNAs.

////  //simple tweedie test
//  @Test public void testTweedieRegression() throws InterruptedException, ExecutionException{
//    Key raw = Key.make("gaussian_test_data_raw");
//    Key parsed = Key.make("gaussian_test_data_parsed");
//    Key<GLMModel> modelKey = Key.make("gaussian_test");
//    Frame fr = null;
//    GLMModel model = null;
//    try {
//      // make data so that the expected coefficients is icept = col[0] = 1.0
//      FVecTest.makeByteVec(raw, "x,y\n0,0\n1,0.1\n2,0.2\n3,0.3\n4,0.4\n5,0.5\n6,0.6\n7,0.7\n8,0.8\n9,0.9\n0,0\n1,0\n2,0\n3,0\n4,0\n5,0\n6,0\n7,0\n8,0\n9,0");
//      fr = ParseDataset.parse(parsed, new Key[]{raw});
//      double [] powers = new double [] {1.5,1.1,1.9};
//      double [] intercepts = new double []{3.643,1.318,9.154};
//      double [] xs = new double []{-0.260,-0.0284,-0.853};
//      for(int i = 0; i < powers.length; ++i){
//        DataInfo dinfo = new DataInfo(fr, 1, false, DataInfo.TransformType.NONE);
//        GLMParameters glm = new GLMParameters(Family.tweedie);
//
//        new GLM2("GLM test of gaussian(linear) regression.",Key.make(),modelKey,dinfo,glm,new double[]{0},0).fork().get();
//        model = DKV.get(modelKey).get();
//        testHTML(model);
//        HashMap<String, Double> coefs = model.coefficients();
//        assertEquals(intercepts[i],coefs.get("Intercept"),1e-3);
//        assertEquals(xs[i],coefs.get("x"),1e-3);
//      }
//    }finally{
//      if( fr != null ) fr.delete();
//      if(model != null)model.delete();
//    }
//  }
@Test
public void testAllNAs() {
    Key raw = Key.make("gamma_test_data_raw");
    Key parsed = Key.make("gamma_test_data_parsed");
    FVecTest.makeByteVec(raw, "x,y,z\n1,0,NA\n2,NA,1\nNA,3,2\n4,3,NA\n5,NA,1\nNA,6,4\n7,NA,9\n8,NA,18\nNA,9,23\n10,31,NA\nNA,11,20\n12,NA,25\nNA,13,37\n14,45,NA\n");
    Frame fr = ParseDataset.parse(parsed, raw);
    GLM job = null;
    try {
        GLMParameters params = new GLMParameters(Family.poisson);
        // params._response = 1;
        params._response_column = fr._names[1];
        params._train = parsed;
        params._lambda = new double[] { 0 };
        params._missing_values_handling = MissingValuesHandling.Skip;
        GLM glm = new GLM(params);
        glm.trainModel().get();
        assertFalse("should've thrown IAE", true);
    } catch (IllegalArgumentException e) {
        assertTrue(e.getMessage().contains("No rows left in the dataset"));
    } finally {
        fr.delete();
    }
}

Example 40 with GLMParameters

use of hex.glm.GLMModel.GLMParameters in project h2o-3 by h2oai.

the class GLMTest method testCoordinateDescent_airlines_CovUpdates.

@Test
public void testCoordinateDescent_airlines_CovUpdates() {
    GLMModel model = null;
    Key parsed = Key.make("airlines_parsed");
    Key<GLMModel> modelKey = Key.make("airlines_model");
    Frame fr = parse_test_file(parsed, "smalldata/airlines/AirlinesTrain.csv.zip");
    try {
        // H2O differs on intercept and race, same residual deviance though
        GLMParameters params = new GLMParameters();
        params._standardize = true;
        params._family = Family.binomial;
        params._solver = Solver.COORDINATE_DESCENT;
        params._response_column = "IsDepDelayed";
        params._ignored_columns = new String[] { "IsDepDelayed_REC" };
        params._train = fr._key;
        GLM glm = new GLM(params, modelKey);
        model = glm.trainModel().get();
        assertTrue(glm.isStopped());
        System.out.println(model._output._training_metrics);
    } finally {
        fr.delete();
        if (model != null)
            model.delete();
    }
}