ignite: how to save and re-load trained model

Question

Following is the piece of code i used to train my model. After that how and where can i save my model and read it back other than FileExporter class? is it only in a file or can i store it in a cache and access back?

IgniteCache<Integer, double[]> cache = ignite.getOrCreateCache("MLData_IRIS");

// extracting   sepal length, sepal width, petal length, petal width
IgniteBiFunction<Integer, double[], Vector> featureExtractor = new RangeExtractor(1, 5);
IgniteBiFunction<Integer, double[], Double> labelExtractor = new PointExtractor(0);

System.out.println(">>> Create new training dataset splitter object.");
TrainTestSplit<Integer, double[]> split = new TrainTestDatasetSplitter<Integer, double[]>()
    .split(0.5, 0.5);

IgniteBiPredicate<Integer, double[]> testData = split.getTestFilter();
IgniteBiPredicate<Integer, double[]> trainData = split.getTrainFilter();

// Set up the trainer
KMeansTrainer trainer = new KMeansTrainer()
    .withDistance(new EuclideanDistance())  //other metrics are HammingDistance, ManhattanDistance
    .withAmountOfClusters(3) // number clusters want to create
    .withMaxIterations(100)
    .withEpsilon(1.0E-4D)
    .withSeed(1234L);

long t1 = System.currentTimeMillis();

KMeansModel mdl = trainer.fit(
    ignite,
    cache,
    trainData,
    featureExtractor,
    labelExtractor
);

long t2 = System.currentTimeMillis();
System.out.println("time taken to build the model : " + (t2 - t1) + " ms");

System.out.println(">>> --------------------------------------------");
System.out.println(">>> trained model: " + mdl.toString(true));

Answer 1

For now Ignite have only this mechanism - FileExporter.

But, for version 2.8 we already implemented model storage.

Sample for saving model:

ModelStorage storage = new ModelStorageFactory().getModelStorage(ignite);
        storage.mkdirs("/");
        storage.putFile("/my_model", serializedMdl);

        ModelDescriptor desc = new ModelDescriptor(
            "MyModel",
            "My Cool Model",
            new ModelSignature("", "", ""),
            new ModelStorageModelReader("/my_model"),
            new IgniteModelParser<>()
        );
        ModelDescriptorStorage descStorage = new ModelDescriptorStorageFactory().getModelDescriptorStorage(ignite);
        descStorage.put("my_model", desc);

Sample for loading model:

Ignite ignite = Ignition.ignite();

        ModelDescriptorStorage descStorage = new ModelDescriptorStorageFactory().getModelDescriptorStorage(ignite);
        ModelDescriptor desc = descStorage.get(mdl);

        Model<byte[], byte[]> infMdl = new SingleModelBuilder().build(desc.getReader(), desc.getParser());

        Vector input = VectorUtils.of(x);

        try {
            return deserialize(infMdl.predict(serialize(input)));
        }
        catch (IOException | ClassNotFoundException e) {
            throw new RuntimeException(e);
        }

Where x - is vector of doubles and mdl - is model name.

NOTE: this API will be available with release 2.8. But, you could try it right now if you will build Ignite from master branch.