- 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.
- 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).
- 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.
- Clarify the Regression interface with
Optional_arg_intf. - Add Cubic spline interpolation.
- Array.zip/unzip
- fst3,snd3,thr3
- 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.
- First public release: ready for other people to start hacking.