Improve GBDT: regression, predict, docs

docs/algo/sona/feature_gbdt_sona.md

@@ -55,42 +55,75 @@ GBDT的训练方法中，核心是一种叫梯度直方图的数据结构，需
 ### 参数
 
 * **算法参数**
-  * ml.num.class：分裂数量
+  * ml.gbdt.task.type：任务类型，分类或者回归
   * ml.gbdt.loss.func：代价函数，支持二分类（binary:logistic）、多分类（multi:logistic）和均方根误差（rmse）
   * ml.gbdt.eval.metric：模型指标，支持rmse、error、log-loss、cross-entropy、precision和auc
+  * ml.num.class：分类数量，仅对分类任务有用
   * ml.gbdt.feature.sample.ratio：特征采样比例（0到1之间）
   * ml.gbdt.tree.num：树的数量	
   * ml.gbdt.tree.depth：树的最大高度
   * ml.gbdt.split.num：每个特征的分裂点的数量
   * ml.learn.rate：学习速率
+  * ml.gbdt.min.node.instance：叶子节点上数据的最少数量
+  * ml.gbdt.min.split.gain：分裂需要的最小增益
+  * ml.gbdt.reg.lambda：正则化系数
+  * ml.gbdt.multi.class.strategy：多分类任务的策略，一轮一棵树(one-tree)或者一轮多棵树(multi-tree)
 
 * **输入输出参数**
 	* angel.train.data.path：训练数据的输入路径
 	* angel.validate.data.path：验证数据的输入路径
+	* angel.predict.data.path：预测数据的输入路径
+	* angel.predict.out.path：预测结果的保存路径
 	* angel.save.model.path：训练完成后，模型的保存路径
+	* angel.load.model.path：预测开始前，模型的加载路径
 
 ### 训练任务启动命令示例
 
 使用spark提交任务
-
+ 
   ./spark-submit \
     --master yarn-cluster \   
     --conf spark.ps.jars=$SONA_ANGEL_JARS \  
-    --jars $SONA_SPARK_JARS  \  
-    --name "LR Adam on Spark-on-Angel" \  
+    --conf spark.ps.cores=1 \
+    --conf spark.ps.memory=10g \
+    --conf spark.ps.log.level=INFO \
+    --queue $queue \
+    --jars $SONA_SPARK_JARS  \
+    --name "GBDT on Spark-on-Angel" \
     --driver-memory 5g \  
     --num-executors 10 \  
     --executor-cores 1 \  
     --executor-memory 10g \   
     --class com.tencent.angel.spark.ml.tree.gbdt.trainer.GBDTTrainer \  
     spark-on-angel-mllib-${ANGEL_VERSION}.jar \   
+    ml.gbdt.task.type:classification \
     angel.train.data.path:XXX angel.validate.data.path:XXX angel.save.model.path:XXX \  
     ml.gbdt.loss.func:binary:logistic ml.gbdt.eval.metric:error,log-loss \  
-    ml.learn.rate:0.1 ml.gbdt.split.num:10  ml.gbdt.tree.num:20 ml.gbdt.tree.depth:7 ml.class.num:2 \ 
-    ml.feature.index.range:47237 ml.gbdt.feature.sample.ratio:1.0
-        
- 
+    ml.learn.rate:0.1 ml.gbdt.split.num:10  ml.gbdt.tree.num:20 ml.gbdt.tree.depth:7 ml.num.class:2 \ 
+    ml.feature.index.range:47237 ml.gbdt.feature.sample.ratio:1.0 ml.gbdt.multi.class.strategy:one-tree ml.gbdt.min.node.instance:100
+
+### 预测任务启动命令示例
 
+使用spark提交任务
+
+  ./spark-submit \
+      --master yarn-cluster \   
+      --conf spark.ps.jars=$SONA_ANGEL_JARS \  
+      --conf spark.ps.cores=1 \
+      --conf spark.ps.memory=10g \
+      --conf spark.ps.log.level=INFO \
+      --queue $queue \
+      --jars $SONA_SPARK_JARS  \
+      --name "GBDT on Spark-on-Angel" \
+      --driver-memory 5g \  
+      --num-executors 10 \  
+      --executor-cores 1 \  
+      --executor-memory 10g \   
+      --class com.tencent.angel.spark.ml.tree.gbdt.predictor.GBDTPredictor \  
+      spark-on-angel-mllib-${ANGEL_VERSION}.jar \
+      angel.load.model.path:XXX angel.predict.data.path:XXX angel.predict.out.path:XXX  \  
+
+
 
 ## 5. 性能
 

spark-on-angel/mllib/src/main/scala/com/tencent/angel/spark/ml/tree/common/TreeConf.java

@@ -25,6 +25,10 @@ public class TreeConf {
   public static final String ML_TRAIN_PATH = "spark.ml.train.path";
   public static final String ML_VALID_PATH = "spark.ml.valid.path";
   public static final String ML_PREDICT_PATH = "spark.ml.predict.path";
+  public static final String ML_OUTPUT_PATH = "spark.ml.output.path";
+
+  public static final String ML_GBDT_TASK_TYPE = "ml.gbdt.task.type";
+  public static final String DEFAULT_ML_GBDT_TASK_TYPE = "classification";
   public static final String ML_VALID_DATA_RATIO = "spark.ml.valid.ratio";
   public static final double DEFAULT_ML_VALID_DATA_RATIO = 0.25;
   public static final String ML_NUM_CLASS = "spark.ml.class.num";

spark-on-angel/mllib/src/main/scala/com/tencent/angel/spark/ml/tree/gbdt/examples/GBDTRegression.scala

@@ -0,0 +1,102 @@
+/*
+ * Tencent is pleased to support the open source community by making Angel available.
+ *
+ * Copyright (C) 2017-2018 THL A29 Limited, a Tencent company. All rights reserved.
+ *
+ * Licensed 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
+ *
+ * https://opensource.org/licenses/Apache-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 com.tencent.angel.spark.ml.tree.gbdt.examples
+
+import com.tencent.angel.spark.ml.core.ArgsUtil
+import com.tencent.angel.spark.ml.tree.gbdt.trainer.GBDTTrainer
+import com.tencent.angel.spark.ml.tree.param.GBDTParam
+import com.tencent.angel.spark.ml.tree.util.Maths
+import org.apache.spark.{SparkConf, SparkContext}
+
+object GBDTRegression {
+
+  def main(args: Array[String]): Unit = {
+
+    @transient val conf = new SparkConf().setMaster("local").setAppName("gbdt")
+
+    val param = new GBDTParam
+
+    // spark conf
+    val numExecutor = 1
+    val numCores = 1
+    param.numWorker = numExecutor
+    param.numThread = numCores
+    conf.set("spark.task.cpus", numCores.toString)
+    conf.set("spark.locality.wait", "0")
+    conf.set("spark.memory.fraction", "0.7")
+    conf.set("spark.memory.storageFraction", "0.8")
+    conf.set("spark.task.maxFailures", "1")
+    conf.set("spark.yarn.maxAppAttempts", "1")
+    conf.set("spark.network.timeout", "1000")
+    conf.set("spark.executor.heartbeatInterval", "500")
+
+    val params = ArgsUtil.parse(args)
+
+    //val trainPath = "data/dna/dna.scale"  //dimension=181
+    //val validPath = "data/dna/dna.scale.t"
+    val trainPath = "data/abalone/abalone_8d_train.libsvm"  //dimension=8
+    val validPath = "data/abalone/abalone_8d_train.libsvm"
+    val modelPath = "tmp/gbdt/abalone"
+
+    // dataset conf
+    param.taskType = "regression"
+    param.numClass = 2
+    param.numFeature = 8
+
+    // loss and metric
+    param.lossFunc = "rmse"
+    param.evalMetrics = Array("rmse")
+    param.multiGradCache = false
+
+    // major algo conf
+    param.featSampleRatio = 1.0f
+    param.learningRate = 0.1f
+    param.numSplit = 10
+    param.numTree = 10
+    param.maxDepth = 7
+    val maxNodeNum = Maths.pow(2, param.maxDepth + 1) - 1
+    param.maxNodeNum = 4096 min maxNodeNum
+
+    // less important algo conf
+    param.histSubtraction = true
+    param.lighterChildFirst = true
+    param.fullHessian = false
+    param.minChildWeight = 0.0f
+    param.minNodeInstance = 10
+    param.minSplitGain = 0.0f
+    param.regAlpha = 0.0f
+    param.regLambda = 1.0f
+    param.maxLeafWeight = 0.0f
+
+    println(s"Hyper-parameters:\n$param")
+
+    @transient implicit val sc = new SparkContext(conf)
+
+    try {
+      val trainer = new GBDTTrainer(param)
+      trainer.initialize(trainPath, validPath)
+      val model = trainer.train()
+      trainer.save(model, modelPath)
+    } catch {
+      case e: Exception =>
+        e.printStackTrace()
+    } finally {
+    }
+  }
+
+}

spark-on-angel/mllib/src/main/scala/com/tencent/angel/spark/ml/tree/gbdt/histogram/Histogram.java

@@ -36,6 +36,7 @@ public Histogram(int numBin, int numClass, boolean fullHessian, boolean multiCla
     this.numClass = numClass;
     this.fullHessian = fullHessian;
     this.multiClassMultiTree = multiClassMultiTree;
+
     if (numClass == 2 || multiClassMultiTree) {
       this.gradients = new double[numBin];
       this.hessians = new double[numBin];