Data Interface

Like the XGBoost python module, XGBoost4J uses DMatrix to handle data.LIBSVM txt format file, sparse matrix in CSR/CSC format, and dense matrix aresupported.

• The first step is to import DMatrix:

  importml.dmlc.xgboost4j.java.DMatrix;

• Use DMatrix constructor to load data from a libsvm text format file:

  DMatrixdmat=newDMatrix("train.svm.txt");

• Pass arrays to DMatrix constructor to load from sparse matrix.

  Suppose we have a sparse matrix

  1 0 2 04 0 0 33 1 2 0

  We can express the sparse matrix inCompressed Sparse Row (CSR) format:

  long[]rowHeaders=newlong[]{0,2,4,7};float[]data=newfloat[]{1f,2f,4f,3f,3f,1f,2f};int[]colIndex=newint[]{0,2,0,3,0,1,2};intnumColumn=4;DMatrixdmat=newDMatrix(rowHeaders,colIndex,data,DMatrix.SparseType.CSR,numColumn);

  … or inCompressed Sparse Column (CSC) format:

  long[]colHeaders=newlong[]{0,3,4,6,7};float[]data=newfloat[]{1f,4f,3f,1f,2f,2f,3f};int[]rowIndex=newint[]{0,1,2,2,0,2,1};intnumRow=3;DMatrixdmat=newDMatrix(colHeaders,rowIndex,data,DMatrix.SparseType.CSC,numRow);

• You may also load your data from a dense matrix. Let’s assume we have a matrix of form

  1    23    45    6

  Usingrow-major layout, we specify the dense matrix as follows:

  float[]data=newfloat[]{1f,2f,3f,4f,5f,6f};intnrow=3;intncol=2;floatmissing=0.0f;DMatrixdmat=newDMatrix(data,nrow,ncol,missing);

• To set weight:

  float[]weights=newfloat[]{1f,2f,1f};dmat.setWeight(weights);

Setting Parameters

To set parameters, parameters are specified as a Map:

Map<String,Object>params=newHashMap<String,Object>(){{put("eta",1.0);put("max_depth",2);put("objective","binary:logistic");put("eval_metric","logloss");}};

Training Model

With parameters and data, you are able to train a booster model.

• Import Booster and XGBoost:

  importml.dmlc.xgboost4j.java.Booster;importml.dmlc.xgboost4j.java.XGBoost;

• Training

  DMatrixtrainMat=newDMatrix("train.svm.txt");DMatrixvalidMat=newDMatrix("valid.svm.txt");// Specify a watch list to see model accuracy on data setsMap<String,DMatrix>watches=newHashMap<String,DMatrix>(){{put("train",trainMat);put("test",testMat);}};intnround=2;Boosterbooster=XGBoost.train(trainMat,params,nround,watches,null,null);

• Saving model

  After training, you can save model and dump it out.

  booster.saveModel("model.json");

• Generating model dump with feature map

  // dump without feature mapString[]model_dump=booster.getModelDump(null,false);// dump with feature mapString[]model_dump_with_feature_map=booster.getModelDump("featureMap.txt",false);

• Load a model

  Boosterbooster=XGBoost.loadModel("model.json");

Prediction

After training and loading a model, you can use it to make prediction for other data. The result will be a two-dimension float array(nsample,nclass); forpredictLeaf(), the result would be of shape(nsample,nclass*ntrees).

DMatrixdtest=newDMatrix("test.svm.txt");// predictfloat[][]predicts=booster.predict(dtest);// predict leaffloat[][]leafPredicts=booster.predictLeaf(dtest,0);