Since I started researching on machine learning and dynamical systems, I have been developing several utility libraries. Most of them are written in modern C++ because of its flexibility and performance.

In this blog, I intend to write my thoughts on machine learning and C++ programming. Specifically, I will discuss the motivation and the design decisions of the C++ libraries I have been developing: Jules and Cool.

Jules

Without question, Jules is my favorite.

I started writing it to overcome several inconveniences I had with other frameworks. My first research experiments (back in 2011) were written in Matlab. At the time, speed and flexibility were not a problem, since the problems were simple. Then, at some point, I decided to try R. I liked the flexibility and the huge amount of available packages which drastically increased my productivity. However, at the beginning of my PhD, I had many problems with the hard maintenance of code. Every time I would update R, I would need to make few (but annoying) changes in the code because of the many dependencies I had. Since I wasn’t happy with the performance either and I had just finished reading the 4th edition of The C++ Programming Language, I decided to write my own framework to deal with statistical programming using modern C++.

Most of my studies comprise only some kind of dynamical system simulation, thus a library with basic array manipulation and statistical utilities (sum, mean, standard deviation, etc) would suffice. There are some other C++ libraries out there that provide such functionalities. Blaze and Eigen are good examples. However, I had some reasons to create yet another (TM) library:

• Modern C++. Jules is not intended to work in old compilers, so I can use the most recent features of C++.
• Productivity over speed. Most of the related C++ libraries are blazing fast and highly configurable, but they are hard to learn and decrease a lot your productivity. Jules, in the contrary, aims at being simple and intuitive, without many configuration options. Moreover, it is header-only, which should help integration in your project. Besides Jules focusing in flexibility and simplicity, it doesn’t mean that it is slow. Soon, I should run some benchmarks and post them here.

Paulo Urio and I started writing it in June 2015. Since then, the library has been evolving in a good pace, taking it in mind that there are only 2 contributors. The library is not stable yet, but I have been using it in every one of my projects. Most of the features I have implemented are directly connected with my needs, but if you intend to use it and need some feature, fill an issue that I will gladly try to implement it.

I will eventually blog about Jules and its usage.

Cool

One problem that I have every time I start a new project is that I see myself reimplementing some simple functionalities. Instead of doing so, or using Boost, I decided to aggregate many simple tools into a single project. Cool aims at providing:

• Easy integration: Every functionality is implemented in a single header file with just a couple hundred lines.
• Productivity over flexibility: Unlike Boost, utilities are not heavily templated or configurable.

Many of my posts here will discuss the rationale and the design decisions of the parts of Cool. I will focus on the usage of modern C++ features.