eidos.ml is a C# / .NET Standard library for productive development of machine learning pipelines that is based on MathNet.Numerics.
It is inspired by scikit-learn but uses advantages of C# syntax and .NET runtime.
The name eidos comes from the Plato's theory of Forms, where it means non-physical (but substantial) forms (or ideas) representing the most accurate reality. Since machine learning models should simulate the reality as accurately as possible an eidos would be an ideal model.
Currently library is an early development stage and suggested use-case is participation in competitions like Kaggle and learning and experimenting with ML algorithms.
There will be breaking changes. But after the release of 1.0 it will be stable and suitable for production for small-scale problems (that fit to a single machine's memory).
- 03.09.2017 Version 0.1 alpha is published with support of basic abstractions, Linear Regression and Ridge Regression.
Why invent the wheel?
There are other great ML libraries for other languages and for C# too. Why create yet another?
Not only for fun. See the corresponding blog post.
Plans for the nearest future
See GitLab issues with tag upcoming and the road-map below.
|0.1||September 2017||other methods for LinearRegression, RidgeRegression, documentation and tests|
|0.2||October 2017||LogisticRegression, common transformations and metrics|
|0.3||November 2017||TBD: most frequently used ML algorithms|
|1.0||2018||stable architecture and backward compatibility, wide range of reliable and fast algorithms|