I have downloaded a pojo from h2o, compiled it, but how do I use it?

Question

I'm using the following sample code to download a pojo that I found from this post:

import h2o
 h2o.init()
 iris_df = h2o.import_file("https://s3.amazonaws.com/h2o-public-test-data/smalldata/iris/iris.csv")
 from h2o.estimators.glm import H2OGeneralizedLinearEstimator
 predictors = iris_df.columns[0:4]
 response_col = "C5"
 train,valid,test = iris_df.split_frame([.7,.15], seed =1234)
 glm_model = H2OGeneralizedLinearEstimator(family="multinomial")
 glm_model.train(predictors, response_col, training_frame = train, validation_frame = valid)
 h2o.download_pojo(glm_model, path = '/Users/your_user_name/Desktop/', get_jar = True)

When I open the downloaded java file I'm given some instructions for how to compile it. The following compiles successfully:

javac -cp h2o-genmodel.jar -J-Xmx2g -J-XX:MaxPermSize=128m GLM_model_python_1488677745392_2.java

Now, I'm not sure how to use it. I've tried the following:

java -cp h2o-genmodel.jar javac -cp h2o-genmodel.jar -J-Xmx2g -J-XX:MaxPermSize=128m GLM_model_python_1488677745392_2.java

The following is the code in the pojo:

/*
  Licensed under the Apache License, Version 2.0
    http://www.apache.org/licenses/LICENSE-2.0.html

  AUTOGENERATED BY H2O at 2017-03-05T01:51:46.237Z
  3.10.3.2

  Standalone prediction code with sample test data for GLMModel named GLM_model_python_1488677745392_2

  How to download, compile and execute:
      mkdir tmpdir
      cd tmpdir
      curl http:/10.0.0.4/10.0.0.4:54321/3/h2o-genmodel.jar > h2o-genmodel.jar
      curl http:/10.0.0.4/10.0.0.4:54321/3/Models.java/GLM_model_python_1488677745392_2 > GLM_model_python_1488677745392_2.java
      javac -cp h2o-genmodel.jar -J-Xmx2g -J-XX:MaxPermSize=128m GLM_model_python_1488677745392_2.java

     (Note:  Try java argument -XX:+PrintCompilation to show runtime JIT compiler behavior.)
*/
import java.util.Map;
import hex.genmodel.GenModel;
import hex.genmodel.annotations.ModelPojo;

@ModelPojo(name="GLM_model_python_1488677745392_2", algorithm="glm")
public class GLM_model_python_1488677745392_2 extends GenModel {
  public hex.ModelCategory getModelCategory() { return hex.ModelCategory.Multinomial; }

  public boolean isSupervised() { return true; }
  public int nfeatures() { return 4; }
  public int nclasses() { return 3; }

  // Names of columns used by model.
  public static final String[] NAMES = NamesHolder_GLM_model_python_1488677745392_2.VALUES;
  // Number of output classes included in training data response column.
  public static final int NCLASSES = 3;

  // Column domains. The last array contains domain of response column.
  public static final String[][] DOMAINS = new String[][] {
    /* C1 */ null,
    /* C2 */ null,
    /* C3 */ null,
    /* C4 */ null,
    /* C5 */ GLM_model_python_1488677745392_2_ColInfo_4.VALUES
  };
  // Prior class distribution
  public static final double[] PRIOR_CLASS_DISTRIB = {0.2818181818181818,0.33636363636363636,0.38181818181818183};
  // Class distribution used for model building
  public static final double[] MODEL_CLASS_DISTRIB = null;

  public GLM_model_python_1488677745392_2() { super(NAMES,DOMAINS); }
  public String getUUID() { return Long.toString(-5598526670666235824L); }

  // Pass in data in a double[], pre-aligned to the Model's requirements.
  // Jam predictions into the preds[] array; preds[0] is reserved for the
  // main prediction (class for classifiers or value for regression),
  // and remaining columns hold a probability distribution for classifiers.
  public final double[] score0( double[] data, double[] preds ) {
    final double [] b = BETA.VALUES;
    for(int i = 0; i < 0; ++i) if(Double.isNaN(data[i])) data[i] = CAT_MODES.VALUES[i];
    for(int i = 0; i < 4; ++i) if(Double.isNaN(data[i + 0])) data[i+0] = NUM_MEANS.VALUES[i];
    preds[0] = 0;
    for(int c = 0; c < 3; ++c){
      preds[c+1] = 0;
      for(int i = 0; i < 4; ++i)
        preds[c+1] += b[0+i + c*5]*data[i];
      preds[c+1] += b[4 + c*5]; // reduce intercept
    }
    double max_row = 0;
    for(int c = 1; c < preds.length; ++c) if(preds[c] > max_row) max_row = preds[c];
    double sum_exp = 0;
    for(int c = 1; c < preds.length; ++c) { sum_exp += (preds[c] = Math.exp(preds[c]-max_row));}
    sum_exp = 1/sum_exp;
    double max_p = 0;
    for(int c = 1; c < preds.length; ++c) if((preds[c] *= sum_exp) > max_p){ max_p = preds[c]; preds[0] = c-1;};
    return preds;
  }
    public static class BETA implements java.io.Serializable {
      public static final double[] VALUES = new double[15];
      static {
        BETA_0.fill(VALUES);
      }
      static final class BETA_0 implements java.io.Serializable {
        static final void fill(double[] sa) {
          sa[0] = -1.4700470387418272;
          sa[1] = 4.26067731522767;
          sa[2] = -2.285756276489862;
          sa[3] = -4.312931422791621;
          sa[4] = 5.231215014401568;
          sa[5] = 1.7769023115830205;
          sa[6] = -0.2534145823550425;
          sa[7] = -0.9887536067536575;
          sa[8] = -1.2706135235877678;
          sa[9] = -4.319817154759757;
          sa[10] = 0.0;
          sa[11] = -3.024835247270209;
          sa[12] = 3.8622405283810464;
          sa[13] = 7.018262604176258;
          sa[14] = -22.702291637028203;
        }
      }
}
// Imputed numeric values
    static class NUM_MEANS implements java.io.Serializable {
      public static final double[] VALUES = new double[4];
      static {
        NUM_MEANS_0.fill(VALUES);
      }
      static final class NUM_MEANS_0 implements java.io.Serializable {
        static final void fill(double[] sa) {
          sa[0] = 5.90272727272727;
          sa[1] = 3.024545454545454;
          sa[2] = 3.9490909090909097;
          sa[3] = 1.2872727272727267;
        }
      }
}
// Imputed categorical values.
    static class CAT_MODES implements java.io.Serializable {
      public static final int[] VALUES = new int[0];
      static {
      }
}
    // Categorical Offsets
    public static final int[] CATOFFS = {0};
}
// The class representing training column names
class NamesHolder_GLM_model_python_1488677745392_2 implements java.io.Serializable {
  public static final String[] VALUES = new String[4];
  static {
    NamesHolder_GLM_model_python_1488677745392_2_0.fill(VALUES);
  }
  static final class NamesHolder_GLM_model_python_1488677745392_2_0 implements java.io.Serializable {
    static final void fill(String[] sa) {
      sa[0] = "C1";
      sa[1] = "C2";
      sa[2] = "C3";
      sa[3] = "C4";
    }
  }
}
// The class representing column C5
class GLM_model_python_1488677745392_2_ColInfo_4 implements java.io.Serializable {
  public static final String[] VALUES = new String[3];
  static {
    GLM_model_python_1488677745392_2_ColInfo_4_0.fill(VALUES);
  }
  static final class GLM_model_python_1488677745392_2_ColInfo_4_0 implements java.io.Serializable {
    static final void fill(String[] sa) {
      sa[0] = "Iris-setosa";
      sa[1] = "Iris-versicolor";
      sa[2] = "Iris-virginica";
    }
  }
}

Now, I think I need to call score0. I've figured out how to create my own main.java and create an entrypoint to main() so that I can instantiate the object and call score0, but I have no idea how it's supposed to work. I'm expecting to feed in 4 doubles and get back a category, but instead, the function takes two double[] and I can't figure out exactly what to put where and how to read the results. Here's my main:

public class Main {
  public static void main(String[] args) {
      double[] input = {4.6, 3.1, 1.5, 0.2};
      double[] output = new double[4];
      GLM_model_python_1488677745392_2 m = new GLM_model_python_1488677745392_2();
      double[] t = m.score0(input,output);
      for(int i = 0; i < t.length; i++) System.out.println(t[i]);
  }
}

I'm actually getting a bunch of data returned, but I don't know what any of it means. I think I'm completely using the second argument incorrectly, but I'm not sure what to do. Here's the output:

0.0
0.9976588811416329
0.0023411188583572825
9.662837354438092E-15

Answer 1

A few points:

• I don't recommend trying to call score0 directly when you are just learning how to use H2O POJOs. The EasyPredictModelWrapper API was created to provide a friendly interface, and I recommend using that if you can. (The best reason to skip the Easy API layer would be if you're interested in pure raw speed.)

• You may find it easier to work with a MOJO instead of a POJO. MOJOs can be used in exactly the same way as POJOs, but they are data representations of the model rather than code representations of the model. MOJOs have strong backwards compatibility guarantees, and do not need to be compiled. I recommend using the MOJO if you can.

• POJO and MOJO online documentation for the latest stable release of H2O can be found here:

  • http://docs.h2o.ai/h2o/latest-stable/h2o-genmodel/javadoc/index.html

• Code for the EasyPredictModelWrapper can be found here:

  • https://github.com/h2oai/h2o-3/blob/master/h2o-genmodel/src/main/java/hex/genmodel/easy/EasyPredictModelWrapper.java

For people that really want to call score0 directly, the best documentation for how to do that is the EasyPredictModelWrapper code.

Here is a POJO usage code snippet from the documentation (of H2O version 3.10.4.1) of how to make a new prediction with the Easy API:

import java.io.*;
import hex.genmodel.easy.RowData;
import hex.genmodel.easy.EasyPredictModelWrapper;
import hex.genmodel.easy.prediction.*;

public class main {
  private static String modelClassName = "gbm_pojo_test";

  public static void main(String[] args) throws Exception {
    hex.genmodel.GenModel rawModel;
    rawModel = (hex.genmodel.GenModel) Class.forName(modelClassName).newInstance();
    EasyPredictModelWrapper model = new EasyPredictModelWrapper(rawModel);
    //
    // By default, unknown categorical levels throw PredictUnknownCategoricalLevelException.
    // Optionally configure the wrapper to treat unknown categorical levels as N/A instead
    // and strings that cannot be converted to numbers also to N/As:
    //
    //     EasyPredictModelWrapper model = new EasyPredictModelWrapper(
    //         new EasyPredictModelWrapper.Config()
    //             .setModel(rawModel)
    //             .setConvertUnknownCategoricalLevelsToNa(true)
    //             .setConvertInvalidNumbersToNa(true)
    //     );

    RowData row = new RowData();
     row.put("Year", "1987");
     row.put("Month", "10");
     row.put("DayofMonth", "14");
     row.put("DayOfWeek", "3");
     row.put("CRSDepTime", "730");
     row.put("UniqueCarrier", "PS");
     row.put("Origin", "SAN");
     row.put("Dest", "SFO");

    BinomialModelPrediction p = model.predictBinomial(row);
    System.out.println("Label (aka prediction) is flight departure delayed: " + p.label);
    System.out.print("Class probabilities: ");
    for (int i = 0; i < p.classProbabilities.length; i++) {
      if (i > 0) {
        System.out.print(",");
      }
      System.out.print(p.classProbabilities[i]);
    }
    System.out.println("");
  }
}

After stuffing a new prediction value into row (which is just a Map), you call predictBinomial() to make a prediction.

Almost exactly the same code can be used for a MOJO, except you need to instantiate the model from a data file instead of from a class. So instead of this code for POJO:

    hex.genmodel.GenModel rawModel;
    rawModel = (hex.genmodel.GenModel) Class.forName(modelClassName).newInstance();
    EasyPredictModelWrapper model = new EasyPredictModelWrapper(rawModel);

you would have this code for MOJO:

    EasyPredictModelWrapper model = new EasyPredictModelWrapper(MojoModel.load("GBM_model_R_1475248925871_74.zip"));

Answer 2

If anyone finds this thread, I found an even easier way to leverage the downloaded pojo. H2o steam handles it nicely:

~/steam-1.1.6-linux-amd64 > java -jar var/master/assets/jetty-runner.jar --port 8888 var/master/assets/ROOT.war &
curl -X POST --form pojo=@/home/tome/pojo/DL_defaults.java --form jar=@/home/tome/pojo/h2o-genmodel.jar localhost:8888/makewar > example.war
java -jar /home/tome/steam-1.1.6-linux-amd64/var/master/assets/jetty-runner.jar --port 7077 example.war

Then you can query it:

01:34:57 PS C:\dropbox\scripts> Invoke-RestMethod "http://notexist.eastus.cloudapp.azure.com:7077/predict?C1=4.6&C2=3.1&C3=1.5&C4=0.2"


labelIndex label       classProbabilities
---------- -----       ------------------
         0 Iris-setosa {0.9976588811416329, 0.0023411188583572825, 9.66283735443809E-15}

The main http: page provides a nice interface for building your query, and if you don't like the above, the full version of Steam provides a way to connect to H2O directly and do the download, conversion, and deployment of a model for you in a couple of clicks.

Answer 3

Never mind - it looks like I had it correct after all. The output refers to the 3 categories I have:

t[1] is setosa
t[2] is versicolor
t[3] is virginica

the 0.9976588811416329 refers to the percentage it thinks it's in that category. So the data is 99.8% setosa.