H2o4gpu Versions Save

• xgboost fix output_margin parameter #834
• Downgrade LightGBM to last stable version(2.2.x)

• ARMA/ARIMA #797
• Random state seed for matrix factorization #803
• Update xgboost while keeping an old version(0.9) to allow convert old pickled model #822
• Update lightgbm
• Remove a lot dependencies and make them less strict #826 #800
• CUDA-10.1 support
• Various bug fixes and dependencies update

• Fix K-means demo for multi-GPU #630
• CUDA-10
• Use gcc to build LightGBM on ppc64le

• Fix memory leak #175 #204
• Static linking with CUDA dependencies
• CUDA forward compatibility
• Update NCCL to 2.4
• Update arrow to 0.12.0
• Update XGBoost
• Update numpy, scipy and scikit-learn
• Add experimental Matrix Factorization #729

• Conda build added
• Power 8 and Power 9 builds
• cuda8, cuda9, cuda92 builds
• Include LightGBM in cpu and gpu modes
• Update XGBoost
• Updates for pygdf/arrow and update associated demo notebooks
• Various bug fixes

• R API
• Intel DAAL initial support
• tSVD improvements

New algorithms:

• Truncated SVD
• PCA

Improvements to existing algorithms:

• KMeans|| for KMeans
• Performance improvements for KMeans
• Performance improvements for XGBoost