Future Plan

What is the future plan of Owl?

Owl aims to be a general OCaml numerical library for scientific computing. The library focuses on the following perspective:

• Dense matrix operations (99% completed)
• Sparse matrix operations (99% completed)
• Advanced mathematical functions (90% completed)
• Advanced statistical functions (70% completed)
• Linear algebra (40% completed)
• Regression functions (40% completed)
• Machine learning (10% completed)
• Basic Plotting functions (75% completed)

As you can see, Owl is still in its infancy phase. I appreciate any comments and look forward to any help from anyone who want to contribute.

What are the specific TODOs?

There is a list of things (either well-defined or not) in my mind to push the project forward. Bold font marks the task I am currently working on.

• Unify all the mathematical operators (long term, need to wait for MI).
• Unify the indexing and slicing operator (long term, need to wait for 4.06)..
• Implement algorithmic differentiation, forward mode.
• Provide a pure OCaml implementation of Sparse Matrix module.
• Support N-dimensional view.
• Add uniform error handling, e.g., check the matrix shape.
• Add hypothesis tests in Stats module.
• A Core module for matrix without dependency on GSL.
• Serialise matrices to standard formats such as HDF5.
• Add integration and ODE functions.
• Add Graph module (build atop of ocamlgraph).
• Extensive tests on Dense matrix module to eliminate bugs.
• Complete some missing functions in Sparse matrix module.
• Make Sparse matrix module support "x.{i,j}" to access the elements.
• Add more types of plots in Plot module.
• Add consistent error management in all modules.
• Complete all the missing functions in Linalg module for dense matrices.
• Automate the document generation using ocamldoc.
• Make a docker image supporting GUI so Plot module can plot in X window.
• Make Linalg module support sparse matrices.
• Implement more optimisation algorithms in Optimise module: e.g., coordinate descent, proximal gradient, and etc.
• Implement more machine learning algorithms (clustering, deep neural networks, and etc.), may introduce a new separate module in Owl.
• Include a set of approximate algorithms in Owl as a separate module (e.g., approximate k-means, approximate k-nn, and etc.)
• Embed a small probabilistic programming language in Stats module.
• Add a web interface so that the plot can be dynamically updated on the fly.
• whatever else helpful ...

What have been DONE?

• Support Float32, Float64, Complex32, Complex64 in Matrix.
• Support sparse N-dimensional array.
• Create project page for owl on github.
• Improve the documentation and write a series of tutorials (posts) for Owl.
• Dense N-dimensional array (Ndarray).
• Add FFT functions.
• Enhance sparse real matrices.
• Add the library to OPAM repository.
• Add the supports of sparse complex matrices.
• Add the supports of dense complex matrices.
• Get rid of ctypes.foreign, use stub generation to interface with GSL
• Add widely used constants to the module, e.g., pi, speed of light, natural number, gravity constant, and etc.