Skip to content

Latest commit

 

History

History
101 lines (79 loc) · 4.83 KB

CHANGELOG.md

File metadata and controls

101 lines (79 loc) · 4.83 KB

[2.0.4] - 2019-04-20

Added

  • linalg: Ported BLAS spr()
  • linalg: Ported Matrix copy() and toString(), and implemented clone() to make matrices usable within RDD functions as a zero value parameter
  • linalg: Ported RowMatrix computeCovariance(), computeGramianMatrix() and triuToFull()
  • util: New utility equivalent to scala.collection.Iterator fill()
  • stat: Ported corr() and PearsonCorrelation class

Changed

  • linalg: Use native Lua 1-based colPtrs indexes in SparseMatrix

Fixed

  • #56 Java array binarySearch() claimed to use 1-based index params and an exclusive end-index, but used 0-based index params and an inclusive end-index

Changed

  • linalg: DenseVector and SparseVector are now clonable, making them usable as a zero value in RDD functions; required by RowMatrix

[2.0.2] - 2019-02-23

Added

  • Document how to train a K-means model within Redis at examples/redis-kmeans
  • Document how to run test suites within Redis at spec-redis
  • #43 Support loading LIBSVM files from MLUtils
  • DenseMatrix and SparseMatrix map() and update() support
  • DenseMatrix diag() and eye() support

Fixed

  • #47 Cannot train a K-means model within Redis due to unnecessary os.date() dependency (Redis interop)

[2.0.1] - 2019-01-16

Added

  • #31 [linalg:] Support DenseMatrix, Matrices, Matrix, and SparseMatrix modules
  • [regression:] Support LabeledPoint class

Changed

  • DenseVector and SparseVector toString() format changed to conform to Apache Spark (instead of Lua conventions), now that LabeledPoint unit test coverage requires compatibility

[2.0.0] - 2018-12-16

Changed

  • Upgraded to Stuart 2.0.0

[1.0.1] - 2018-11-17

Fixed

  • #38 Vectors.dense(...) with varargs error: no table.pack() function (eLua interop)

[1.0.0] - 2018-11-08

Changed

  • Upgrade to Stuart 1.0.0 with changes to class framework

[0.1.9] - 2018-11-02

Added

  • #28 Train a K-means model

Changed

  • #34 Remove all use of module upvalues, so that modules can be more easily transpiled to C then burned into ROM and chipsets (eLua LTR interop)

Fixed

  • #23 K-means cluster centers load in a non-deterministic sort order due to missing sort
  • #25 VectorWithNorm initializer fails to compute norm when not explicitly provided

[0.1.8] - 2018-10-14

Added

  • Lua 5.3 support
  • #17 stat: Ported MultivariateOnlineSummarizer class, RowMatrix numCols(), numRows(), computeColumnSummaryStatistics(), and updateNumRows(), and statistics colStats()
  • #20 Package and deploy releases to npmjs.com and jsDelivr

Changed

  • #15 Remove stuart-sql LuaRocks dependency. It is still used when present, but no longer required.

[0.1.7] - 2018-10-12

Changed

  • Upgrade to Stuart 0.1.7

[0.1.5] - 2017-12-12

Added

  • clustering: Ported KMeansModel
  • util: Ported Loader, which can load a KMeansModel from a Parquet file or directory of files, whether local or WebHDFS
  • util: Ported NumericParser
  • util: Ported StringTokenizer

[0.1.3] - 2017-11-11

Added

  • clustering: Ported KMeans fastSquaredDistance(), findClosest() and pointCost()
  • linalg: Ported BLAS axpy(), dot(), and scal()
  • linalg: Ported Vectors sqdist()
  • linalg: Implement VectorWithNorm equality (__eq)
  • util: Ported MLUtils EPSILON and fastSquaredDistance()

Changed

  • clustering: Fixed KMeans getters and setters to match RDD API, not DataFrame API
  • linalg: Vector types are now 0-based like Apache Spark
  • Consolidate Apache Spark and Stuart unit tests into a single unified folder hierarchy

Fixed

  • linalg: Vector numActives and numNonzeros fields were not updating after changes to vector. BLAS functions mutate vectors. Changed fields to functions so that they're computed each time.

[0.1.0] - 2017-10-28

Added

  • clustering: Ported VectorWithNorm datatype
  • linalg: Ported DenseVector and SparseVector datatypes, plus their Vectors factory

(formatted per keepachangelog-1.1.0)