[WIP] Reinforcement learning experiments (Dyna-Q+) using game theory (network routing games). A rather straight-forward/simple directed graph is used to assess the performance.
LightGraphs.jl: SimpleWeightedGraph is the base graph type. Algorithms such as A* or the recursive depth-first search (DFS) are used.
JuMP.jl: the linear programming (LP) optimization framework is used to solve a min-cost flow problem (optimal assignment).
Test.jl and Logging.jl modules are used for unit testing the code and output debug information. Run:
julia maintest.jl
Code coverage is still missing.
There's a lot of back and forth on benchmarking Python, C++ and Julia. In my opinion, Python is probably the best scripting language for quick&dirty prototyping. When it comes to experimenting, MATLAB is my preferred option. However, as everyone claims, Julia lies in between Python and the gold standard for high performance computing, C++.
When I was thinking about doing an abstraction of the problem, I realized Julia was:
- more handy,
- easy-to-interpret,
- as abstract-rich as a pure OOP language, and
- with a very nice LaTeX typesetting/interpreter in Atom.
[1] Roughgarden, T. (2004). Stackelberg scheduling strategies. SIAM journal on computing, 33(2), 332-350.
[2] Sutton, R. S., & Barto, A. G. (2018). Reinforcement learning: An introduction. MIT press.
[3] Kwon, C. (2016). Julia programming for operations research: a primer on computing.