IGNITE-11039: [ML] Add parser for Spark Decision tree regression
authorzaleslaw <zaleslaw.sin@gmail.com>
Thu, 31 Jan 2019 16:05:18 +0000 (19:05 +0300)
committerYury Babak <ybabak@gridgain.com>
Thu, 31 Jan 2019 16:05:18 +0000 (19:05 +0300)
This closes #5987

examples/src/main/java/org/apache/ignite/examples/ml/inference/spark/modelparser/DecisionTreeRegressionFromSparkExample.java [new file with mode: 0644]
examples/src/main/resources/models/spark/serialized/dtreg/data/._SUCCESS.crc [new file with mode: 0644]
examples/src/main/resources/models/spark/serialized/dtreg/data/.part-00000-366f6ff2-698b-4bdd-8b1c-de87e11b3d1b-c000.snappy.parquet.crc [new file with mode: 0644]
examples/src/main/resources/models/spark/serialized/dtreg/data/_SUCCESS [new file with mode: 0644]
examples/src/main/resources/models/spark/serialized/dtreg/data/part-00000-366f6ff2-698b-4bdd-8b1c-de87e11b3d1b-c000.snappy.parquet [new file with mode: 0644]
examples/src/main/resources/models/spark/serialized/dtreg/metadata/._SUCCESS.crc [new file with mode: 0644]
examples/src/main/resources/models/spark/serialized/dtreg/metadata/.part-00000.crc [new file with mode: 0644]
examples/src/main/resources/models/spark/serialized/dtreg/metadata/_SUCCESS [new file with mode: 0644]
examples/src/main/resources/models/spark/serialized/dtreg/metadata/part-00000 [new file with mode: 0644]
modules/ml/spark-model-parser/src/main/java/org/apache/ignite/ml/sparkmodelparser/SparkModelParser.java
modules/ml/spark-model-parser/src/main/java/org/apache/ignite/ml/sparkmodelparser/SupportedSparkModels.java

diff --git a/examples/src/main/java/org/apache/ignite/examples/ml/inference/spark/modelparser/DecisionTreeRegressionFromSparkExample.java b/examples/src/main/java/org/apache/ignite/examples/ml/inference/spark/modelparser/DecisionTreeRegressionFromSparkExample.java
new file mode 100644 (file)
index 0000000..f09cc0c
--- /dev/null
@@ -0,0 +1,91 @@
+/*
+ * 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.spark.modelparser;
+
+import java.io.FileNotFoundException;
+import javax.cache.Cache;
+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.tutorial.TitanicUtils;
+import org.apache.ignite.ml.math.functions.IgniteBiFunction;
+import org.apache.ignite.ml.math.primitives.vector.Vector;
+import org.apache.ignite.ml.math.primitives.vector.VectorUtils;
+import org.apache.ignite.ml.sparkmodelparser.SparkModelParser;
+import org.apache.ignite.ml.sparkmodelparser.SupportedSparkModels;
+import org.apache.ignite.ml.tree.DecisionTreeNode;
+
+/**
+ * Run Decision tree regression model loaded from snappy.parquet file.
+ * The snappy.parquet file was generated by Spark MLLib model.write.overwrite().save(..) operator.
+ * <p>
+ * You can change the test data used in this example and re-run it to explore this algorithm further.</p>
+ */
+public class DecisionTreeRegressionFromSparkExample {
+    /** Path to Spark Decision tree regression model. */
+    public static final String SPARK_MDL_PATH = "examples/src/main/resources/models/spark/serialized/dtreg/data" +
+        "/part-00000-366f6ff2-698b-4bdd-8b1c-de87e11b3d1b-c000.snappy.parquet";
+
+    /** Run example. */
+    public static void main(String[] args) throws FileNotFoundException {
+        System.out.println();
+        System.out.println(">>> Decision tree regression model loaded from Spark through serialization over partitioned 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, Object[]> dataCache = TitanicUtils.readPassengers(ignite);
+
+            IgniteBiFunction<Integer, Object[], Vector> featureExtractor = (k, v) -> {
+                double[] data = new double[] {(double)v[0], (double)v[1], (double)v[5], (double)v[6]};
+                data[0] = Double.isNaN(data[0]) ? 0 : data[0];
+                data[1] = Double.isNaN(data[1]) ? 0 : data[1];
+                data[2] = Double.isNaN(data[2]) ? 0 : data[2];
+                data[3] = Double.isNaN(data[3]) ? 0 : data[3];
+                return VectorUtils.of(data);
+            };
+
+            IgniteBiFunction<Integer, Object[], Double> lbExtractor = (k, v) -> (double)v[4];
+
+            DecisionTreeNode mdl = (DecisionTreeNode)SparkModelParser.parse(
+                SPARK_MDL_PATH,
+                SupportedSparkModels.DECISION_TREE_REGRESSION
+            );
+
+            System.out.println(">>> Decision tree regression model: " + mdl);
+
+            System.out.println(">>> ---------------------------------");
+            System.out.println(">>> | Prediction\t| Ground Truth\t|");
+            System.out.println(">>> ---------------------------------");
+
+            try (QueryCursor<Cache.Entry<Integer, Object[]>> observations = dataCache.query(new ScanQuery<>())) {
+                for (Cache.Entry<Integer, Object[]> observation : observations) {
+                    Vector inputs = featureExtractor.apply(observation.getKey(), observation.getValue());
+                    double groundTruth = lbExtractor.apply(observation.getKey(), observation.getValue());
+                    double prediction = mdl.predict(inputs);
+
+                    System.out.printf(">>> | %.4f\t\t| %.4f\t\t|\n", prediction, groundTruth);
+                }
+            }
+
+            System.out.println(">>> ---------------------------------");
+        }
+    }
+}
diff --git a/examples/src/main/resources/models/spark/serialized/dtreg/data/._SUCCESS.crc b/examples/src/main/resources/models/spark/serialized/dtreg/data/._SUCCESS.crc
new file mode 100644 (file)
index 0000000..3b7b044
Binary files /dev/null and b/examples/src/main/resources/models/spark/serialized/dtreg/data/._SUCCESS.crc differ
diff --git a/examples/src/main/resources/models/spark/serialized/dtreg/data/.part-00000-366f6ff2-698b-4bdd-8b1c-de87e11b3d1b-c000.snappy.parquet.crc b/examples/src/main/resources/models/spark/serialized/dtreg/data/.part-00000-366f6ff2-698b-4bdd-8b1c-de87e11b3d1b-c000.snappy.parquet.crc
new file mode 100644 (file)
index 0000000..8c92f70
Binary files /dev/null and b/examples/src/main/resources/models/spark/serialized/dtreg/data/.part-00000-366f6ff2-698b-4bdd-8b1c-de87e11b3d1b-c000.snappy.parquet.crc differ
diff --git a/examples/src/main/resources/models/spark/serialized/dtreg/data/_SUCCESS b/examples/src/main/resources/models/spark/serialized/dtreg/data/_SUCCESS
new file mode 100644 (file)
index 0000000..e69de29
diff --git a/examples/src/main/resources/models/spark/serialized/dtreg/data/part-00000-366f6ff2-698b-4bdd-8b1c-de87e11b3d1b-c000.snappy.parquet b/examples/src/main/resources/models/spark/serialized/dtreg/data/part-00000-366f6ff2-698b-4bdd-8b1c-de87e11b3d1b-c000.snappy.parquet
new file mode 100644 (file)
index 0000000..0d55ef9
Binary files /dev/null and b/examples/src/main/resources/models/spark/serialized/dtreg/data/part-00000-366f6ff2-698b-4bdd-8b1c-de87e11b3d1b-c000.snappy.parquet differ
diff --git a/examples/src/main/resources/models/spark/serialized/dtreg/metadata/._SUCCESS.crc b/examples/src/main/resources/models/spark/serialized/dtreg/metadata/._SUCCESS.crc
new file mode 100644 (file)
index 0000000..3b7b044
Binary files /dev/null and b/examples/src/main/resources/models/spark/serialized/dtreg/metadata/._SUCCESS.crc differ
diff --git a/examples/src/main/resources/models/spark/serialized/dtreg/metadata/.part-00000.crc b/examples/src/main/resources/models/spark/serialized/dtreg/metadata/.part-00000.crc
new file mode 100644 (file)
index 0000000..e3f446f
Binary files /dev/null and b/examples/src/main/resources/models/spark/serialized/dtreg/metadata/.part-00000.crc differ
diff --git a/examples/src/main/resources/models/spark/serialized/dtreg/metadata/_SUCCESS b/examples/src/main/resources/models/spark/serialized/dtreg/metadata/_SUCCESS
new file mode 100644 (file)
index 0000000..e69de29
diff --git a/examples/src/main/resources/models/spark/serialized/dtreg/metadata/part-00000 b/examples/src/main/resources/models/spark/serialized/dtreg/metadata/part-00000
new file mode 100644 (file)
index 0000000..d6e7670
--- /dev/null
@@ -0,0 +1 @@
+{"class":"org.apache.spark.ml.regression.DecisionTreeRegressionModel","timestamp":1548250179470,"sparkVersion":"2.2.0","uid":"dtr_db43f37d09f7","paramMap":{"featuresCol":"features","minInstancesPerNode":1,"predictionCol":"prediction","checkpointInterval":10,"cacheNodeIds":false,"maxBins":32,"minInfoGain":0.0,"impurity":"variance","maxDepth":7,"labelCol":"age","maxMemoryInMB":256,"seed":926680331},"numFeatures":4}
index 9e8a28c..8a66a3c 100644 (file)
@@ -87,11 +87,21 @@ public class SparkModelParser {
                 return loadRandomForestModel(ignitePathToMdl);
             case KMEANS:
                 return loadKMeansModel(ignitePathToMdl);
+            case DECISION_TREE_REGRESSION:
+                return loadDecisionTreeRegressionModel(ignitePathToMdl);
             default:
                 throw new UnsupportedSparkModelException(ignitePathToMdl);
         }
     }
 
+    /**
+     * Load Decision Tree Regression model.
+     *
+     * @param pathToMdl Path to model.
+     */
+    private static Model loadDecisionTreeRegressionModel(String pathToMdl) {
+        return loadDecisionTreeModel(pathToMdl);
+    }
 
     private static Model loadKMeansModel(String pathToMdl) {
         Vector[] centers = null;
index f5ee3a6..4b203fe 100644 (file)
@@ -41,6 +41,9 @@ public enum SupportedSparkModels {
     /** K-Means. */
     KMEANS,
 
+    /** Decision tree regression. */
+    DECISION_TREE_REGRESSION,
+
     /**
      * Gradient boosted trees.
      * NOTE: support binary classification only with raw labels 0 and 1