Skip to content

CHANGELOG.md

Latest commit

 

History

History
217 lines (181 loc) · 7.39 KB

CHANGELOG.md

File metadata and controls

217 lines (181 loc) · 7.39 KB
Raw

Change Log

  • 4.0.0

    • Data are now backed by a contiguous C buffer instead of PHP array
    • Added fromArray() factory method to build from PHP array
    • build() and quick() factory methods are now deprecated
    • Tensor constructors are no longer public
    • Added dynamic kernel dispatching based on CPU architecture
    • Added AVX/AVX2 and AVX512 element-wise and unary kernels
    • Implemented 2-slot huge buffer memory pinning
    • Modulus results no longer rounded to nearest integer
    • Standardized serial representation with extension
    • Rename Special interface to Reductions and Algebraic to Unary
    • Moved clip(), clipLower(), and clipUpper() into Unary interface
    • Remove poisson factory method
    • Fused covariance() into a single BLAS syrk kernel
    • Add Eigen::eigenvaluesImaginary() exposes imaginary part of each eigenvalue
    • Matrix::eig() no longer discards imaginary part of complex eigenvalue pairs
    • Convolutions now use tiled kernels with FMA and vectorization
    • Added sigmoid() to Unary interface and tanh() to Trigonometric
    • Added softmax() as a fused kernel in the new Special interface
    • Moved Statistical interface into Reductions interface
    • Added padding argument to convolve()

  • 3.1.2

    • Fix extension name

  • 3.1.1

    • Just triggering the first Packagist release

  • 3.1.0

  • 3.0.11

    • Update to Zephir 1.5.0
    • Fixed RREF/REF pivot float tolerance
    • RREF now zeroes the tail if below EPSILON
    • rank(), fullRank(), and inverse() are now consistent
    • Matrix::inverse() now throws a RuntimeException on singular input
    • Matrix::inverse() now throws an InvalidArgumentException on non-square input
    • Fix Poisson distribution when lambda is zero
    • Added square check to inverse()
    • Cholesky decompose() now throws a RuntimeException on non-positive-definite input
    • Added the SingularMatrix exception
    • The extension now reduces singular matrices to REF and RREF instead of throwing

  • 3.0.10

    • Update to Zephir 0.23, adds support for PHP 8.5
    • Fixed Matrix and Vector build() validation
    • Fixed bug in Matrix row variance
    • Fixed LU decomposition pivot selection, singular matrices now throw a RuntimeException
    • Fixed undefined array key in Vector and Matrix quantile at q=1.0 or with a single element
    • Fix extension rank() operation
    • Fixed reduce() callback argument order inconsistency between Matrix and Vector
    • Fix covariance() summation axis
    • Fixed Vector notEqualMatrix throwing InvalidArgumentException instead of DimensionalityMismatch
    • Fixed memory leak in the extension on the failure path of inverse, pseudoinverse

  • 3.0.7

    • Upgrade to Zephir version 0.19
    • Drop support for PHP 7.4 in extension

  • 3.0.6

    • Upgrade to Zephir version 0.18

  • 3.0.5

    • Upgraded Zephir dependency to 0.17

  • 3.0.4

    • Fixed bug in Vector less equal operation

  • 3.0.3

    • Added PHP 8.2 extension compatibility

  • 3.0.2

    • Added PHP 8.1 extension compatibility

  • 3.0.1

    • Add positive semi-definite typehinting

  • 3.0.0

    • No changes

  • 3.0.0-beta

    • Add support for PHP 8.0 in the extension
    • Drop extension support for PHP 7.2 and 7.3
    • Tensors only compute floating point operations
    • Remove dependency on JAMA library
    • Remove previously deprecated items
    • Matrix returns vector on row access
    • Added Algebraic and Special function interfaces

  • 2.2.3

    • Remove dependency on pthreads

-2.2.2 - Compensate for PSR-4 issues

  • 2.2.1

    • Optimize 2D convolution operation
    • Fix and optimize vector convolve 1D operation

  • 2.2.0

    • Implemented Singular Value Decomposition (SVD)
    • Implemented matrix (Moore-Penrose) pseudoinverse
    • Tensor objects now throw namespaced exceptions
    • Optimized Reduced Row Echelon Form (RREF)
    • Optimize eigendecomposition for symmetric matrices

  • 2.1.4

    • Implemented Eigen decomposition in extension
    • Optimize row echelon form (REF) of matrix
    • Optimize Cholesky decomposition
    • Fix memory leak in matrix inverse
    • Optimize LU decomposition

  • 2.1.3

    • Use BLAS DGEMM for matrix multiplication
    • Optimize matrix inverse

  • 2.1.2

    • Optimize scalar arithmetic and comparisons

  • 2.1.1

    • Zephir redux
    • Blanket extension optimizations
    • Added array arithmetic and comparison functions

  • 2.1.0

    • Extension now compiles directly from C source code
    • Remove Zephir extension code
    • Update instructions for extension compilation using PHPize

  • 2.0.5

    • Percentile method changed to quantile
    • Fixed Vector linspace precision

  • 2.0.4

    • Individual Arithmetic and Comparison methods now public
    • Matrix reductions have their own namespace
    • Optimized Matrix structural operations
    • Added argmin and argmax custom optimizers
    • Added matrix/sub matrix insertion

  • 2.0.3

    • Matrix decomp methods now return decomp objects
    • Optimized matrix statistical operations
    • Optimized vector outer product
    • Optimized Matrix flatten
    • Cannot instantiate null/empty tensors

  • 2.0.2

    • Optimized random number generation

  • 2.0.1

    • Added benchmarks
    • Matrix stacking now row or column-wise
    • Changed method signature and behavior of repeat matrix

  • 2.0.0

    • Implemented the library as an extension using Zephir lang
    • Changed namespace from Rubix\Tensor to \Tensor
    • Matrix decompositions are now a separate abstraction
    • Removed Dimensionality Mismatch exception
    • Added inverse trigonometric methods to Trigonometric interface
    • Added Array-Like interface
    • Removed Column/Row exclude methods from Matrix
    • Changed method signature of matrix determinant
    • Added return sub matrix
    • Added positive definite and semi-definite methods to Matrix
    • Added is symmetric method to Matrix
    • Added Statistical and Trigonometric interfaces
    • Added Arithmetic and Comparable interfaces
    • Added log1p and expm1 methods to the Arithmetic interface
    • Added matrix full rank method
    • Added generate random Poisson distribution
    • Added Cholesky decomposition
    • Variance now takes an optional mean argument

  • 1.0.4

    • Added transpose to Tensor API
    • Reduced memory footprint of matmul operation
    • Removed magic getters
    • Added shape string method to Tensor API
    • Improved error messages for matrix dimensionality mismatch

  • 1.0.3

    • Added clip upper and lower bounds
    • Added isSquare method to Matrix
    • Added vector late static binding

  • 1.0.2

    • Added ref using row elimination method
    • Added universal comparison methods to tensor API
    • Added convolve operation to Vector and Matrix

  • 1.0.1

    • Added Column Vector
    • Implemented Eigenvalue decomposition
    • Added solve system of linear equations
    • Integration with 3rd party JAMA library
    • Fixed variance covariance calculation
    • Added percentile calculation for Vector and Matrix

  • 1.0.0

    • Implemented LU decomposition
    • Added vector projection
    • Implemented Matrix inverse
    • Implemented Row Echelon and Reduced Row Echelon decomposition
    • Added Matrix determinant
    • Added Matrix rank
    • Added Matrix/Vector products
    • Implemented universal element-wise tensor operations
    • Added statistical functions
    • Added trigonometric functions
    • Added exponential and logarithmic functions
    • Added tensor factories