Improved performance of CSV Data Source, Low Order Moments, initialization stage of implicit ALS, K-means for CSR datasets for the processors supporting Intel® Advanced Vector Extensions 2 (Intel® AVX2) and Intel® Advanced Vector Extensions 512 (Intel® AVX-512
LBFGS algorithm now supports automatic step-length selection on each iteration of this algorithm by line search to satisfy the Wolfe conditions if specific parameters were set.
Batch size b and correction pair batch size b_H are equal to the number of observations in the training set.
The L parameter is set to 1.
The α^tstep-length sequence is not specified.
Introduced support for MT2203 random number generators. Decision forest APIs changes that include:
API to pass an instance of random number generator to algorithm.
API to retrieve updated instance of random number generator after training stage that can be used in other computations.
Introduced limited FreeBSD support in DAAL Open Source version.
Windows version of DAAL static libraries are Universal Windows driver* compatible.