The story so far…

MicroGP1

Unreleased

The study of an evolutionary tool for generating real assembly programs started around 2000. The first operational tool dates back to 2001: it was composed of a few hundred lines of C code, plus a collection of scripts hacked together to test the feasibility of the idea (DOI: 10.1109/CEC.2002.1004462). It has been retroactively labelled MicroGP1, although no specific name was used at the time.

MicroGP1 internally stores an assembly program as a directed acyclic graphs, where each node encodes a fragment of code with some parameters. Evolution proceeds by tweaking the topology of the graph_manager and modifying the parameters inside the nodes. The graph_manager is eventually transformed into a valid assembly-language program in order to be evaluated, that is, assembled, linked and executed.

MicroGP2

Copyright (c) 2002-06 Giovanni Squillero; distributed under GPL2; available on GitHub.

As its predecessor, MicroGP2 internally encodes programs as graphs (directed multigraphs, this time), but it allows to load the list of the parametric code fragments (macros) from a file, targeting a specific microprocessor (hence the micro in the name). The evolutionary core has been completely re-designed looking at the Genetic Programming paradigm (hence the GP in the name).

MicroGP2 was originally coded in C in a frenzy week while recovering from varicella, and the first release was named Chicken Pox. The initial development has been supported by Intel through a grant named “GP Based Test Program generation”; the final release (April Fool) consists of about 15,000 lines. It is described in the paper MicroGP — An Evolutionary Assembly Program Generator (DOI: 10.1007/s10710-005-2985-x).

Over the years, MicroGP2 was mainly exploited by engineers for the test and verification of small microprocessors, but it was also able to scale up tackling the post-silicon validation of a real Pentium 4 (DOI: 10.1109/MTV.2004.5). In a far less mundane research, it created the first machine-written programs ever able to become King of the Hill in all the main international core-war competitions (DOI: 10.1109/TEVC.2005.856207).

With time, MicroGP2 has been coerced into solving problems it was not meant for, such as design of the bayesian networks, the creation of mathematical functions represented as trees, integer and combinatorial optimization, real-value parameter optimization. Ultimately it was re-implemented from scratch in C++.

MicroGP3

Copyright (c) 2006-16 Giovanni Squillero; distributed under GPL3; available on both GitHub and Sourceforge.

MicroGP3 brought a complete change of paradigm: the focus goes from the problem to the tool, and the main design goal shifts from the solution of a specific class of problems to the development of a tool that can tackle a wide range of possible applications.

The third epoch of development started in 2006 (release: Noodle Soup) with the intent to provide a clean implementation able to replicate the behavior of its predecessors. The subsequent releases were Bluebell (2010), described in the book Evolutionary Optimization: the µGP toolkit (DOI: 10.1007/978-0-387-09426-7); Mistletoe (2012); PalmTree (2014); and Camellia (2015). The coding of Thuban was eventually discontinued in 2016.

Byron

Copyright (c) 2023-24 Giovanni Squillero and Alberto Tonda; distributed under Apache-2.0; available on GitHub.

Byron is a reboot of the MicroGP project: The tool has been re-designed from scratch in Python to take advantage of the peculiar features of the language and of its huge standard library. Byron further increases usability thanks to state-of-the-art cooperative platforms: the tool is available as a PyPi package; and the full documentation, sooner or later, is going to be hosted on Read the Docs.