Release notes

0.3.20

Make the fix in PR backwards compatible in order to use it in H2O-3 and provide same predictions for older MOJOs and pass the compatibility test. PR

0.3.19

Change the order of the floating point operation to match prediction in native xgboost version 1.3.0 and newer PR

0.3.18

Add support for reg:logistic and rank:ndcg oobjectives. PR

0.3.17

Revert renaming of getWeight. PR

0.3.16

Expose RegTreeNode stats. PR

0.3.15

Fix loading an empty gblinear booster PR

0.3.14

Upgrade to XGBoost v1.0.0 support. PR1, PR2

0.3.0

#27 Support DART model.

0.2.1

Support an objective function: "rank:pairwise"

0.2.0

Support XGBoost4J-Spark-generated model file format.
Introduce xgboost-predictor-spark.

0.1.8

Make Predictor Spark-friendly (implement Serializable interface, #11 )

0.1.7

Support latest model file format.
- Commit log of xgboost

0.1.6

Improve the speed performance of prediction:
- Optimize tree retrieval performance.

0.1.5

Support an objective function: "reg:linear"

0.1.4

Improve the speed performance of prediction:
- Introduce methods Predictor#predictSingle() for predicting single value efficiently.

0.1.3

Improve the speed performance of prediction:
- Use Jafama for calculating sigmoid function faster.
- Calling ObjFunction.useFastMathExp(true) you can use Jafama's FastMath.exp().

0.1.2

#2 Add linear models (GBLinear).

0.1.1

#1 Allow users to register their ObjFunction.

0.1.0

Initial release.