examples/src/main/java/org/apache/ignite/examples/ml/multiclass/OneVsRestClassificationExample.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.multiclass;

import java.io.IOException;
import java.util.Arrays;
import javax.cache.Cache;
import org.apache.commons.math3.util.Precision;
import org.apache.ignite.Ignite;
import org.apache.ignite.IgniteCache;
import org.apache.ignite.Ignition;
import org.apache.ignite.cache.query.QueryCursor;
import org.apache.ignite.cache.query.ScanQuery;
import org.apache.ignite.examples.ml.util.MLSandboxDatasets;
import org.apache.ignite.examples.ml.util.SandboxMLCache;
import org.apache.ignite.ml.dataset.feature.extractor.impl.DummyVectorizer;
import org.apache.ignite.ml.math.primitives.vector.Vector;
import org.apache.ignite.ml.multiclass.MultiClassModel;
import org.apache.ignite.ml.multiclass.OneVsRestTrainer;
import org.apache.ignite.ml.preprocessing.Preprocessor;
import org.apache.ignite.ml.preprocessing.minmaxscaling.MinMaxScalerTrainer;
import org.apache.ignite.ml.svm.SVMLinearClassificationModel;
import org.apache.ignite.ml.svm.SVMLinearClassificationTrainer;

/**
 * Run One-vs-Rest multi-class classification trainer ({@link OneVsRestTrainer}) parametrized by binary SVM classifier
 * ({@link SVMLinearClassificationTrainer}) over distributed dataset to build two models: one with min-max scaling and
 * one without min-max scaling.
 * <p>
 * Code in this example launches Ignite grid and fills the cache with test data points (preprocessed
 * <a href="https://archive.ics.uci.edu/ml/datasets/Glass+Identification">Glass dataset</a>).</p>
 * <p>
 * After that it trains two One-vs-Rest multi-class models based on the specified data - one model is with min-max
 * scaling and one without min-max scaling.</p>
 * <p>
 * Finally, this example loops over the test set of data points, applies the trained models to predict what cluster does
 * this point belong to, compares prediction to expected outcome (ground truth), and builds
 * <a href="https://en.wikipedia.org/wiki/Confusion_matrix">confusion matrix</a>.</p>
 * <p>
 * You can change the test data used in this example and re-run it to explore this algorithm further.</p> NOTE: the
 * smallest 3rd class could not be classified via linear SVM here.
 */
public class OneVsRestClassificationExample {
    /**
     * Run example.
     */
    public static void main(String[] args) throws IOException {
        System.out.println();
        System.out.println(">>> One-vs-Rest SVM Multi-class classification model over cached dataset usage example started.");
        // Start ignite grid.
        try (Ignite ignite = Ignition.start("examples/config/example-ignite.xml")) {
            System.out.println(">>> Ignite grid started.");

            IgniteCache<Integer, Vector> dataCache = null;
            try {
                dataCache = new SandboxMLCache(ignite).fillCacheWith(MLSandboxDatasets.GLASS_IDENTIFICATION);

                OneVsRestTrainer<SVMLinearClassificationModel> trainer
                    = new OneVsRestTrainer<>(new SVMLinearClassificationTrainer()
                    .withAmountOfIterations(20)
                    .withAmountOfLocIterations(50)
                    .withLambda(0.2)
                    .withSeed(1234L)
                );

                MultiClassModel<SVMLinearClassificationModel> mdl = trainer.fit(
                    ignite,
                    dataCache,
                    new DummyVectorizer<Integer>().labeled(0)
                );

                System.out.println(">>> One-vs-Rest SVM Multi-class model");
                System.out.println(mdl.toString());

                MinMaxScalerTrainer<Integer, Vector> minMaxScalerTrainer = new MinMaxScalerTrainer<>();

                Preprocessor<Integer, Vector> preprocessor = minMaxScalerTrainer.fit(
                    ignite,
                    dataCache,
                    new DummyVectorizer<Integer>().labeled(0)
                );

                MultiClassModel<SVMLinearClassificationModel> mdlWithScaling = trainer.fit(
                    ignite,
                    dataCache,
                    preprocessor
                );

                System.out.println(">>> One-vs-Rest SVM Multi-class model with MinMaxScaling");
                System.out.println(mdlWithScaling.toString());

                System.out.println(">>> ----------------------------------------------------------------");
                System.out.println(">>> | Prediction\t| Prediction with MinMaxScaling\t| Ground Truth\t|");
                System.out.println(">>> ----------------------------------------------------------------");

                int amountOfErrors = 0;
                int amountOfErrorsWithMinMaxScaling = 0;
                int totalAmount = 0;

                // Build confusion matrix. See https://en.wikipedia.org/wiki/Confusion_matrix
                int[][] confusionMtx = {{0, 0, 0}, {0, 0, 0}, {0, 0, 0}};
                int[][] confusionMtxWithMinMaxScaling = {{0, 0, 0}, {0, 0, 0}, {0, 0, 0}};

                try (QueryCursor<Cache.Entry<Integer, Vector>> observations = dataCache.query(new ScanQuery<>())) {
                    for (Cache.Entry<Integer, Vector> observation : observations) {
                        Vector val = observation.getValue();
                        Vector inputs = val.copyOfRange(1, val.size());
                        double groundTruth = val.get(0);

                        double prediction = mdl.predict(inputs);
                        double predictionWithMinMaxScaling = mdlWithScaling.predict(inputs);

                        totalAmount++;

                        // Collect data for model
                        if (!Precision.equals(groundTruth, prediction, Precision.EPSILON))
                            amountOfErrors++;

                        int idx1 = (int)prediction == 1 ? 0 : ((int)prediction == 3 ? 1 : 2);
                        int idx2 = (int)groundTruth == 1 ? 0 : ((int)groundTruth == 3 ? 1 : 2);

                        confusionMtx[idx1][idx2]++;

                        // Collect data for model with min-max scaling
                        if (!Precision.equals(groundTruth, predictionWithMinMaxScaling, Precision.EPSILON))
                            amountOfErrorsWithMinMaxScaling++;

                        idx1 = (int)predictionWithMinMaxScaling == 1 ? 0 : ((int)predictionWithMinMaxScaling == 3 ? 1 : 2);
                        idx2 = (int)groundTruth == 1 ? 0 : ((int)groundTruth == 3 ? 1 : 2);

                        confusionMtxWithMinMaxScaling[idx1][idx2]++;

                        System.out.printf(">>> | %.4f\t\t| %.4f\t\t\t\t\t\t| %.4f\t\t|\n", prediction, predictionWithMinMaxScaling, groundTruth);
                    }
                    System.out.println(">>> ----------------------------------------------------------------");
                    System.out.println("\n>>> -----------------One-vs-Rest SVM model-------------");
                    System.out.println("\n>>> Absolute amount of errors " + amountOfErrors);
                    System.out.println("\n>>> Accuracy " + (1 - amountOfErrors / (double)totalAmount));
                    System.out.println("\n>>> Confusion matrix is " + Arrays.deepToString(confusionMtx));

                    System.out.println("\n>>> -----------------One-vs-Rest SVM model with MinMaxScaling-------------");
                    System.out.println("\n>>> Absolute amount of errors " + amountOfErrorsWithMinMaxScaling);
                    System.out.println("\n>>> Accuracy " + (1 - amountOfErrorsWithMinMaxScaling / (double)totalAmount));
                    System.out.println("\n>>> Confusion matrix is " + Arrays.deepToString(confusionMtxWithMinMaxScaling));

                    System.out.println(">>> One-vs-Rest SVM model over cache based dataset usage example completed.");
                }
            }
            finally {
                if (dataCache != null)
                    dataCache.destroy();
            }
        }
        finally {
            System.out.flush();
        }
    }
}