Bulk Operation

SPARK-30641 ml.optim.aggregator avoid re-allocating buffers

SPARK-30641 optimize sparse GEMM by skipping bound checking

SPARK-30641 add a common softmax function

SPARK-30641 add new gemv to skip array shape checking

SPARK-30641 Refactor LinearRegression - make huber support virtual centering

SPARK-30641 Refactor AFT - support virtual centering

SPARK-30641 Refactor LinearSVC - support virtual centering

SPARK-30641 Multinomial Logistic Regression with intercept support centering

SPARK-30641 Binary Logistic Regression with intercept support centering

SPARK-30641 Refactor Logistic Aggregator - support virtual centering

SPARK-30641 Linear Models standardization optimization

SPARK-30641 adaptively blockify instances

SPARK-30641 potential regression if use memoryUsage instead of numRows

SPARK-30641 Huber loss Convergence

SPARK-30641 use MemoryUsage to control the size of block

SPARK-30641 Performance test on dense and sparse datasets

SPARK-30641 Performance test on java vectorization vs dot vs gemv vs gemm

SPARK-30641 Document usage of blockSize

SPARK-30641 AFT blockify input vectors

SPARK-30641 GMM blockify input vectors

SPARK-30641 ALS/MLP extend HasBlockSize

SPARK-30641 KMeans blockify input vectors

SPARK-30641 LinearRegression blockify input vectors

SPARK-30641 LogisticRegression blockify input vectors

SPARK-30641 LinearSVC blockify input vectors