Examples with DeepLearningParameters hex.deeplearning.DeepLearningModel.DeepLearningParameters used on opensource projects

Example 41 with DeepLearningParameters

use of hex.deeplearning.DeepLearningModel.DeepLearningParameters in project h2o-3 by h2oai.

the class DeepLearningTest method elasticAveraging.

@Ignore
@Test
public void elasticAveraging() {
    DeepLearningParameters dl;
    Frame frTrain;
    int N = 2;
    DeepLearningModel[] models = new DeepLearningModel[N];
    dl = new DeepLearningParameters();
    Scope.enter();
    //new Random().nextBoolean();
    boolean covtype = true;
    if (covtype) {
        frTrain = parse_test_file("./smalldata/covtype/covtype.20k.data");
        Vec resp = frTrain.lastVec().toCategoricalVec();
        frTrain.remove(frTrain.vecs().length - 1).remove();
        frTrain.add("Response", resp);
    } else {
        frTrain = parse_test_file("./bigdata/server/HIGGS.csv");
        Vec resp = frTrain.vecs()[0].toCategoricalVec();
        frTrain.remove(0).remove();
        frTrain.prepend("Response", resp);
    }
    DKV.put(frTrain);
    try {
        for (int i = 0; i < N; ++i) {
            dl._train = frTrain._key;
            String[] n = ((Frame) DKV.getGet(dl._train)).names();
            if (covtype) {
                dl._response_column = n[n.length - 1];
                dl._ignored_columns = null;
            } else {
                dl._response_column = n[0];
                dl._ignored_columns = new String[] { n[22], n[23], n[24], n[25], n[26], n[27], n[28] };
            }
            dl._export_weights_and_biases = true;
            dl._hidden = new int[] { 64, 64 };
            dl._quiet_mode = false;
            dl._max_w2 = 10;
            dl._l1 = 1e-5;
            dl._reproducible = false;
            //every node only has a piece of the data
            dl._replicate_training_data = false;
            //use multi-node
            dl._force_load_balance = true;
            dl._epochs = 10;
            //100 M/R steps per epoch
            dl._train_samples_per_iteration = frTrain.numRows() / 100;
            dl._elastic_averaging = i == 1;
            dl._elastic_averaging_moving_rate = 0.999;
            dl._elastic_averaging_regularization = 1e-4;
            // Invoke DL and block till the end
            DeepLearning job = new DeepLearning(dl);
            // Get the model
            models[i] = job.trainModel().get();
        }
        for (int i = 0; i < N; ++i) {
            if (models[i] != null)
                Log.info(models[i]._output._training_metrics.cm().table().toString());
        }
        if (models[0] != null)
            Log.info("Without elastic averaging: error=" + models[0]._output._training_metrics.cm().err());
        if (models[1] != null)
            Log.info("With elastic averaging:    error=" + models[1]._output._training_metrics.cm().err());
    //      if (models[0] != null && models[1] != null)
    //        Assert.assertTrue(models[1]._output._training_metrics.cm().err() < models[0]._output._training_metrics.cm().err());
    } finally {
        frTrain.remove();
        for (int i = 0; i < N; ++i) if (models[i] != null)
            models[i].delete();
        Scope.exit();
    }
}

Example 42 with DeepLearningParameters

use of hex.deeplearning.DeepLearningModel.DeepLearningParameters in project h2o-3 by h2oai.

the class DeepLearningTest method testVarimp.

@Test
public void testVarimp() {
    Frame tfr = null;
    DeepLearningModel dl = null;
    try {
        tfr = parse_test_file("./smalldata/iris/iris.csv");
        DeepLearningParameters parms = new DeepLearningParameters();
        parms._train = tfr._key;
        parms._epochs = 100;
        parms._response_column = "C5";
        parms._reproducible = true;
        parms._classification_stop = 0.7;
        parms._score_duty_cycle = 1;
        parms._score_interval = 0;
        parms._hidden = new int[] { 100, 100 };
        parms._seed = 0xdecaf;
        parms._variable_importances = true;
        // Build a first model; all remaining models should be equal
        dl = new DeepLearning(parms).trainModel().get();
        Assert.assertTrue(dl.varImp()._varimp != null);
        //for code coverage only
        Log.info(dl.model_info().toStringAll());
        //all features matter
        Assert.assertTrue(ArrayUtils.minValue(dl.varImp()._varimp) > 0.5);
        //all features matter
        Assert.assertTrue(ArrayUtils.maxValue(dl.varImp()._varimp) <= 1);
    } finally {
        if (tfr != null)
            tfr.delete();
        if (dl != null)
            dl.delete();
    }
}

Example 43 with DeepLearningParameters

use of hex.deeplearning.DeepLearningModel.DeepLearningParameters in project h2o-3 by h2oai.

the class DeepLearningTest method testCheckpointOverwriteWithBestModel.

// Check that the restarted model honors overwrite_with_best_model
@Test
public void testCheckpointOverwriteWithBestModel() {
    Frame tfr = null;
    DeepLearningModel dl = null;
    DeepLearningModel dl2 = null;
    Frame train = null, valid = null;
    try {
        tfr = parse_test_file("./smalldata/iris/iris.csv");
        FrameSplitter fs = new FrameSplitter(tfr, new double[] { 0.8 }, new Key[] { Key.make("train"), Key.make("valid") }, null);
        fs.compute2();
        train = fs.getResult()[0];
        valid = fs.getResult()[1];
        DeepLearningParameters parms = new DeepLearningParameters();
        parms._train = train._key;
        parms._valid = valid._key;
        parms._epochs = 1;
        parms._response_column = "C5";
        parms._reproducible = true;
        parms._hidden = new int[] { 50, 50 };
        parms._seed = 0xdecaf;
        parms._train_samples_per_iteration = 0;
        parms._score_duty_cycle = 1;
        parms._score_interval = 0;
        parms._stopping_rounds = 0;
        parms._overwrite_with_best_model = true;
        dl = new DeepLearning(parms).trainModel().get();
        double ll1 = ((ModelMetricsMultinomial) dl._output._validation_metrics).logloss();
        DeepLearningParameters parms2 = (DeepLearningParameters) parms.clone();
        parms2._epochs = 10;
        parms2._checkpoint = dl._key;
        dl2 = new DeepLearning(parms2).trainModel().get();
        double ll2 = ((ModelMetricsMultinomial) dl2._output._validation_metrics).logloss();
        Assert.assertTrue(ll2 <= ll1);
    } finally {
        if (tfr != null)
            tfr.delete();
        if (dl != null)
            dl.delete();
        if (dl2 != null)
            dl2.delete();
        if (train != null)
            train.delete();
        if (valid != null)
            valid.delete();
    }
}

Example 44 with DeepLearningParameters

use of hex.deeplearning.DeepLearningModel.DeepLearningParameters in project h2o-3 by h2oai.

the class DeepLearningTest method testCategoricalEncodingEigen.

@Test
public void testCategoricalEncodingEigen() {
    Frame tfr = null;
    Frame vfr = null;
    DeepLearningModel dl = null;
    try {
        String response = "survived";
        tfr = parse_test_file("./smalldata/junit/titanic_alt.csv");
        vfr = parse_test_file("./smalldata/junit/titanic_alt.csv");
        if (tfr.vec(response).isBinary()) {
            Vec v = tfr.remove(response);
            tfr.add(response, v.toCategoricalVec());
            v.remove();
        }
        if (vfr.vec(response).isBinary()) {
            Vec v = vfr.remove(response);
            vfr.add(response, v.toCategoricalVec());
            v.remove();
        }
        DKV.put(tfr);
        DKV.put(vfr);
        DeepLearningParameters parms = new DeepLearningParameters();
        parms._train = tfr._key;
        parms._valid = vfr._key;
        parms._response_column = response;
        parms._reproducible = true;
        parms._hidden = new int[] { 20, 20 };
        parms._seed = 0xdecaf;
        parms._categorical_encoding = Model.Parameters.CategoricalEncodingScheme.Eigen;
        parms._score_training_samples = 0;
        dl = new DeepLearning(parms).trainModel().get();
        Assert.assertEquals(((ModelMetricsBinomial) dl._output._training_metrics)._logloss, ((ModelMetricsBinomial) dl._output._validation_metrics)._logloss, 1e-8);
    } finally {
        if (tfr != null)
            tfr.remove();
        if (vfr != null)
            vfr.remove();
        if (dl != null)
            dl.delete();
    }
}

Example 45 with DeepLearningParameters

use of hex.deeplearning.DeepLearningModel.DeepLearningParameters in project h2o-3 by h2oai.

the class DeepLearningTest method testLossFunctions.

// just a simple sanity check - not a golden test
@Test
public void testLossFunctions() {
    Frame tfr = null, vfr = null, fr2 = null;
    DeepLearningModel dl = null;
    for (DeepLearningParameters.Loss loss : new DeepLearningParameters.Loss[] { DeepLearningParameters.Loss.Automatic, DeepLearningParameters.Loss.Quadratic, DeepLearningParameters.Loss.Huber, DeepLearningParameters.Loss.Absolute, DeepLearningParameters.Loss.Quantile }) {
        Scope.enter();
        try {
            tfr = parse_test_file("smalldata/glm_test/cancar_logIn.csv");
            for (String s : new String[] { "Merit", "Class" }) {
                Scope.track(tfr.replace(tfr.find(s), tfr.vec(s).toCategoricalVec()));
            }
            DKV.put(tfr);
            DeepLearningParameters parms = new DeepLearningParameters();
            parms._train = tfr._key;
            parms._epochs = 1;
            parms._reproducible = true;
            parms._hidden = new int[] { 50, 50 };
            parms._response_column = "Cost";
            parms._seed = 0xdecaf;
            parms._loss = loss;
            // Build a first model; all remaining models should be equal
            DeepLearning job = new DeepLearning(parms);
            dl = job.trainModel().get();
            ModelMetricsRegression mm = (ModelMetricsRegression) dl._output._training_metrics;
            if (loss == DeepLearningParameters.Loss.Automatic || loss == DeepLearningParameters.Loss.Quadratic)
                Assert.assertEquals(mm._mean_residual_deviance, mm._MSE, 1e-6);
            else
                assertTrue(mm._mean_residual_deviance != mm._MSE);
            assertTrue(dl.testJavaScoring(tfr, fr2 = dl.score(tfr), 1e-5));
        } finally {
            if (tfr != null)
                tfr.remove();
            if (dl != null)
                dl.delete();
            if (fr2 != null)
                fr2.remove();
            Scope.exit();
        }
    }
}