package oml

Version 0.0.7 (2017-02-21):

• Reworked separation of library in Oml (OCaml only) and Oml_full (includes C/Fortran source links), such that the former is a functional (types and modules work together) subset of the latter.
• Split up the various interfaces and changed some type names used to describe classification and regression.
• Implemented discrete KL divergence.

Version 0.0.6 (2016-08-19):

• Alias method for sampling from Categorical distribution and correcting multinomial to categorical (Sergei Lebedev @superbobry).
• Create build for Oml_lite.
• Remove the joiner tool to rely exclusively on Ocamlbuild.
• Median and Mean absolute deviation.
• Cosine distance.
• Classification via linear (and Quadratic) discriminant analysis.
• Benford's law (Carmelo Piccione @structured).

Version 0.0.5 (2016-02-03):

• Refine Logistic Regression to tune gradient calculation in L-BFGS.

  • Changed to a more general softmax algorithm.
  • Split apart binary vs multiclass.
  • MNIST training script.

• Finish exposing regression capabilities.

  • Coefficient test and stastics.
• Add Dirichlet.
• Running -> Online now functorized.
• Move datasets out of Oml

• Hierarchical reorganization into 6 principal packs.

  • Cleaned up the 'optional' argument logic, now all of these types are contained in the module/functor that uses them.
  • Build system changed for testing and documentation logic. No more joiner. mli's are custom generated for subpacks.

Version 0.0.3 (2015-10-14):

• Clarify the Regression interface with Optional_arg_intf.
• Add Cubic spline interpolation.
• Array.zip/unzip
• fst3,snd3,thr3

Version 0.0.2 (2015-09-09):

• Clean up the Classify API via a functor.
• Clean up test suite to use bounded floats and have a simpler testing framework.
• Add bisection method.

Version 0.0.1 (2015-08-06):

• First public release: ready for other people to start hacking.