RELEASENOTES.md

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.