Export - CSV (All fields)
Export - CSV (Current fields)
SYSTEMDS-2204
Missing deregistration of parallelized RDDs on cleanup
RemoveEmpty empty.return parameter ignored during runtime
KMeans w/ codegen shows very bad performance
Regression scoring failing w/ enabled codegenSYSTEMDS-2196
Multi-threaded RDD parallelize
Use common pool in multi-threaded operations
Abstraction for common thread pool
Unnecessary evictions on rdd collect
Double check new single-precision native conv2d on different platforms
Stratstats performance issues for 1M x 1K due to fallback to cpmm
Stratstats performance issues for 100K x 1K due to evictions
Stratstats performance issues for 10K x 1K due to wdivmm
Lost backwards compatibility w/ Spark 2.1 and 2.2
Univariate stats over 10Kx1K w/ wrong optimizer choiceSYSTEMDS-2181
Perftest for SystemML 1.1 release
Fuse all heuristic failing for kmeans tests w/o rewrites
Update janino version from 3.0.0 to 3.0.8
Update ANTLR version from 4.5.3 to 4.7.1 to match Spark 2.3
Set spark cpmm preferred join-multiply parallelism
Performance ultra-sparse reshape, especially colwise vector-matrix
Ultra-sparse cpmm empty block pruning
Upper bound nnz propagation over spark instructions
Improved caching decision ultra-sparse matrices with invalid cp dims
Remote parfor fails on reading ultra-sparse matrix with dims > 2G
Add support for nary rbind operations
Reversed print ordering after function inlining
Skip redundant read/deserialization of shared dictionary
Performance large partitioned broadcasts
Deserialization of compressed blocks w/ shared dictionary fails in Spark
Suboptimal codegen plans for ALS due to wrong cost model memoization
Incorporate sparsity in lower bound costs
Size propagation issue nary rbind
Codegen optimizer row to cell conversion destroys fusion plans
Side effects of parfor recompilation on codegen plans, e.g., kmeans
Recompute nnz on spark checkpoints for large matrices
Misleading explain recompile_hop output
Parfor optimizer fails checking for block-partitioned inputs
Spark mapmm fails for broadcast with larger than integer dims
Codegen produced incorrect results for sparse linregCG
Codegen optimizer creates suboptimal plan for L2SVM over mnist80m
New rewrite for value replacements
Performance cell operations over compressed matrices
Performance read of short and wide sparse matrices
Kmeans w/ codegen fails on distributed spark operations
Codegen counter-productive w/ forced GPU ops
Unsupported sprop/sigmoid in codegen row templates
Spark matrix reshape hangs on result index creation
Codegen fails on comiling ifelse with scalar predicate
Text to binary block conversions fails for very wide, ultra-sparse data
Support matrices with zero rows/columns