public abstract static class SharedTreeModel.SharedTreeOutput
extends hex.Model.Output
Modifier and Type | Field and Description |
---|---|
double |
_init_f
InitF value (for zero trees)
f0 = mean(yi) for gaussian
f0 = log(yi/1-yi) for bernoulli
For GBM bernoulli, the initial prediction for 0 trees is
p = 1/(1+exp(-f0))
From this, the mse for 0 trees can be computed as follows:
mean((yi-p)^2)
This is what is stored in _scored_train[0]
|
int |
_ntrees
Number of trees actually in the model (as opposed to requested)
|
hex.ScoreKeeper[] |
_scored_train |
hex.ScoreKeeper[] |
_scored_valid |
long[] |
_training_time_ms
Training time
|
water.Key<CompressedTree>[][] |
_treeKeys
Trees get big, so store each one seperately in the DKV.
|
TreeStats |
_treeStats
More indepth tree stats
|
water.util.TwoDimTable |
_variable_importances
Variable importances computed during training
|
hex.VarImp |
_varimp |
Constructor and Description |
---|
SharedTreeModel.SharedTreeOutput(SharedTree b,
double mse_train,
double mse_valid) |
Modifier and Type | Method and Description |
---|---|
void |
addKTrees(DTree[] trees) |
CompressedTree |
ctree(int tnum,
int knum) |
java.lang.String |
toStringTree(int tnum,
int knum) |
addModelMetrics, classNames, getModelCategory, hasOffset, hasWeights, isClassifier, isSupervised, nclasses, nfeatures, offsetIdx, offsetName, responseIdx, responseName, toString, weightsIdx, weightsName
public double _init_f
public int _ntrees
public final TreeStats _treeStats
public water.Key<CompressedTree>[][] _treeKeys
public hex.ScoreKeeper[] _scored_train
public hex.ScoreKeeper[] _scored_valid
public long[] _training_time_ms
public water.util.TwoDimTable _variable_importances
public hex.VarImp _varimp
public SharedTreeModel.SharedTreeOutput(SharedTree b, double mse_train, double mse_valid)
public void addKTrees(DTree[] trees)
public CompressedTree ctree(int tnum, int knum)
public java.lang.String toStringTree(int tnum, int knum)