examples/src/main/java/org/apache/ignite/examples/ml/inference/TensorFlowDistributedInferenceExample.java

/*
 * Licensed to the Apache Software Foundation (ASF) under one or more
 * contributor license agreements.  See the NOTICE file distributed with
 * this work for additional information regarding copyright ownership.
 * The ASF licenses this file to You under the Apache License, Version 2.0
 * (the "License"); you may not use this file except in compliance with
 * the License.  You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

package org.apache.ignite.examples.ml.inference;

import java.io.File;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Collections;
import java.util.List;
import java.util.Objects;
import java.util.Random;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.Future;
import org.apache.ignite.Ignite;
import org.apache.ignite.Ignition;
import org.apache.ignite.internal.util.IgniteUtils;
import org.apache.ignite.ml.inference.Model;
import org.apache.ignite.ml.inference.builder.IgniteDistributedModelBuilder;
import org.apache.ignite.ml.inference.parser.ModelParser;
import org.apache.ignite.ml.inference.parser.TensorFlowSavedModelModelParser;
import org.apache.ignite.ml.inference.reader.FileSystemModelReader;
import org.apache.ignite.ml.inference.reader.ModelReader;
import org.apache.ignite.ml.util.MnistUtils;
import org.tensorflow.Tensor;

/**
 * This example demonstrates how to: load TensorFlow model into Java, make inference in distributed environment using
 * Apache Ignite services.
 */
public class TensorFlowDistributedInferenceExample {
    /**
     * Path to the directory with saved TensorFlow model.
     */
    private static final String MODEL_PATH = "examples/src/main/resources/models/mnist_tf_model";

    /**
     * Path to the MNIST images data.
     */
    private static final String MNIST_IMG_PATH = "examples/src/main/resources/datasets/t10k-images-idx3-ubyte";

    /**
     * Path to the MNIST labels data.
     */
    private static final String MNIST_LBL_PATH = "examples/src/main/resources/datasets/t10k-labels-idx1-ubyte";

    /**
     * Run example.
     */
    public static void main(String[] args) throws IOException, ExecutionException, InterruptedException {
        try (Ignite ignite = Ignition.start("examples/config/example-ignite.xml")) {
            File mdlRsrc = IgniteUtils.resolveIgnitePath(MODEL_PATH);
            if (mdlRsrc == null)
                throw new IllegalArgumentException("Resource not found [resource_path=" + MODEL_PATH + "]");

            ModelReader reader = new FileSystemModelReader(mdlRsrc.getPath());

            ModelParser<double[], Long, ?> parser = new TensorFlowSavedModelModelParser<double[], Long>("serve")

                .withInput("Placeholder", doubles -> {
                    float[][][] reshaped = new float[1][28][28];
                    for (int i = 0; i < doubles.length; i++)
                        reshaped[0][i / 28][i % 28] = (float)doubles[i];
                    return Tensor.create(reshaped);
                })

                .withOutput(Collections.singletonList("ArgMax"), collectedTensors -> collectedTensors.get("ArgMax")
                    .copyTo(new long[1])[0]);

            List<MnistUtils.MnistLabeledImage> images = MnistUtils.mnistAsList(
                Objects.requireNonNull(IgniteUtils.resolveIgnitePath(MNIST_IMG_PATH)).getPath(),
                Objects.requireNonNull(IgniteUtils.resolveIgnitePath(MNIST_LBL_PATH)).getPath(),
                new Random(0),
                10000
            );

            long t0 = System.currentTimeMillis();

            try (Model<double[], Future<Long>> threadedMdl = new IgniteDistributedModelBuilder(ignite, 4, 4)
                .build(reader, parser)) {
                List<Future<?>> futures = new ArrayList<>(images.size());
                for (MnistUtils.MnistLabeledImage image : images)
                    futures.add(threadedMdl.predict(image.getPixels()));
                for (Future<?> f : futures)
                    f.get();
            }

            long t1 = System.currentTimeMillis();

            System.out.println("Threaded model throughput: " + images.size() / ((t1 - t0) / 1000.0) + " req/sec");
        }
        finally {
            System.out.flush();
        }
    }
}