R h2o load a saved model from disk in MOJO or POJO format

Question

I\'m catching up on h2o\'s MOJO and POJO model format. I\'m able to save a model in MOJO/POJO with

h2o.download_mojo(model, path = \"/media/somewhe         


        

Answer 1

h2o.loadModel is meant to be used with h2o.saveModel. If you want to compile and run a MOJO you need to do the following:

first let's say you created a MOJO from a GBM:

library(h2o)
h2o.init(nthreads=-1)
path = "http://h2o-public-test-data.s3.amazonaws.com/smalldata/prostate/prostate.csv"
h2o_df = h2o.importFile(path)
h2o_df$RACE = as.factor(h2o_df$RACE)
model = h2o.gbm(y="CAPSULE",
        x=c("AGE", "RACE", "PSA", "GLEASON"),
        training_frame=h2o_df,
        distribution="bernoulli",
        ntrees=100,
        max_depth=4,
        learn_rate=0.1)

and then downloaded the MOJO and the resulting h2o-genmodel.jar file to a new experiment folder. Note that the h2o-genmodel.jar file is a library that supports scoring and contains the required readers and interpreters. This file is required when MOJO models are deployed to production.

modelfile = model.download_mojo(path="~/experiment/", get_genmodel_jar=True)
print("Model saved to " + modelfile)
Model saved to /Users/user/GBM_model_R_1475248925871_74.zip"

Then you would open a new terminal window and change into the experiment directory where you have have the MOJO files .zip and .jar.

$ cd experiment

Then you would create your main program in the experiment folder by creating a new file called main.java (for example, using "vim main.java"). Include the following contents. Note that this file is referencing the GBM model created above using R.

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

public class main {
  public static void main(String[] args) throws Exception {
    EasyPredictModelWrapper model = new EasyPredictModelWrapper(MojoModel.load("GBM_model_R_1475248925871_74.zip"));

    RowData row = new RowData();
    row.put("AGE", "68");
    row.put("RACE", "2");
    row.put("DCAPS", "2");
    row.put("VOL", "0");
    row.put("GLEASON", "6");

    BinomialModelPrediction p = model.predictBinomial(row);
    System.out.println("Has penetrated the prostatic capsule (1=yes; 0=no): " + 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("");
  }
}

Then compile and run in terminal window 2 to get a display of predicted probabilities

$ javac -cp h2o-genmodel.jar -J-Xms2g -J-XX:MaxPermSize=128m main.java
$ java -cp .:h2o-genmodel.jar main  

Answer 2

If you are looking to make predictions on an H2O model in R, then you have three options (which method you choose depends on your use-case):

1. You can use a binary model instead of a MOJO (or POJO). For this method, you export the model to disk using h2o.saveModel() and load it back into the H2O clsuter using h2o.loadModel() and make predictions using predict(model, test). This method requires having an H2O cluster running.
2. If you's still prefer to export a model to MOJO (or POJO) format, you can use the h2o.mojo_predict_df() or h2o.mojo_predict_csv() function in R to generate predictions on a test set (from an R data.frame or in a CSV file).
3. As an alternative to #2, if your data is in JSON format, you can use h2o.predict_json(), but it will only score one row at a time.

Answer 3

Newer versions of H2O have the ability to import MOJOs via the python API:

# re-import saved MOJO
imported_model = h2o.import_mojo(path)

new_observations = h2o.import_file(path='new_observations.csv')
predictions = imported_model.predict(new_observations)

Caution: MOJO cannot be re-imported into python in older H2O versions, which lack the h2o.import_mojo() function.

So h2o.save_model() seems to have lost its role - we can use just my_model.save_mojo() (notice it's not a h2o method, but a property of the model object), as these files can be used not just for Java apps deployment, but also in python as well (in fact they still use a python-Java bridge for that internally).