The V Machine Learning Roadmap and Ecosystem

[ulises-jeremias]

This issue defines a first iteration of the Machine Learning Roadmap. All the specific tasks will be linked to each corresponding repository and also contained on the ML Project here.

We can think a ML Ecosystem as a composition of five different categories: Array Protocols, Foundation, Technique Specific, Domain-Specific and Application Specific.

Array Protocols

For this category we seek to have an alternative implementation to arrays in V. In V it is not possible to create a structure that uses the same operators as an array, so it is not possible to implement a structure that has the same protocol as an array.

However, we started working on VTL, a library similar to numpy, Arraymancer, among others that offers a Ndarray (in this case called Tensor).

Foundation

On this category I identify two aspects: Algorithms and Plotting tools.
On this area we did a cool progress and we already have some libraries for this:

• V Scientific Library, VSL., is a library with a great variety of different modules. Although most modules offer pure-V definitions, VSL also provides modules that wrap known C libraries among other backends that allow high performance computing as an alternative.
• vsl.plot is the plot submodule of VSL that follows the structure of
  Plotly's graph_objects to plot data. Will provide more backends in the future and also a pure V implementation at some point.

Thirdparty

• erdetn.vplot is a wrapper for GNU Plot (gnuplot_i)

Technique Specific

• vsl.ml has Kmeans, LinearRegression, KNN, Params Regression, NLP, and a Data wrapper with an observer pattern.
• . . .

Thirdparty

• hamnn is a machine learning library for classification using a nearest neighbor algorithm based on Hamming distances.
• vervet is a Dataframe manipulation tool written in V

Domain Specific

TODO

Application Specific

TODO

[danieldaeschle]

A keras/tf implementation would also be nice!

[ulises-jeremias]

@danieldaeschle sure! an initial iteration of that will be covered when https://github.com/vlang/vtl/tree/main/nn gets released! it is still WIP

[ArtemkaKun]

@ulises-jeremias Hi, maybe we should close/update/move this issue?

[trufae]

What about adding bindings for the c api exposed from llama2.cpp?

[joe-conigliaro]

Imo a very important topic, since python is a clear leader in ML space due to the existing libraries (and ease of use). I think there is a great opportunity for another language to make a name for itself.

BTW have you seen https://github.com/tinygrad/tinygrad ?

[ulises-jeremias]

yes, was one of my inspirations on VTL 😊 just need more time/help on it

[ulises-jeremias]

will update this before the Hacktoberfest. Also will add issues to VSL and VTL to continue working adding functionalities for ML. Right now working on VTL.nn