naclsn/grom

What if a video-game ROM was also a genome in a genetic algorithm?

grom

Overview

grom defines a Genome class meant to be loaded from a file as an array of bytes. You can then apply modification function like mutate or crossover to edit the Genome or generate a new Genome. Finally you can use save to save into a file, then start to run the file with the associated program.

The same algorithms will works regardless of the data itself: the file you use may vary from .gb to plain .exe or even sources files like .png...

---

The Genome

The Genome is an object built around a bytearray to ease its loading, saving and modification:

from grom import Genome

# load the content of "test/filename.ext" into a `bytearray`
g = Genome("test/filename.ext", "genome/output.ext")

To a loaded Genome, you can apply the following data modification function:

g.mutate(.001, 1) # apply a random mutation of ampliture 1 to .1% of the data

g.geneswap(10, 8) # swap 10 random chunks of 8 bytes in the data

The crossover function works a bit differently as it does not modify the Genome itself, but rather create a new one:

g2 = Genom("another/filename.ext")

# crossover `g` and `g2` into a new `Genome`
child = g.crossover(g2, "child/filename.ext")

You may then save the Genome to use with its associated program:

child.save() # save the child into "child/filename.ext"
child.start() # start the file using Python's `os.startfile`

# or simply:
child()

Note that almost every function (all function that does not return something in particular) returns self to allow for chaining:

Genome(filename).partition(parts).mutate(tx, amp).save().start()

---

Partitioning

You can define a partition system to facilitate gene modification, like local mutations. For that, the Genome object will behave like an array of delimited and named spaces, attributed to a range in the data.

Let's take classic .bmp file:

1. The first 0x36 bytes are reserved for signature and header information
2. Bytes from 0x36 to EOF contains the raw pixel data we want to modify

Note: header size may vary..?

You may then want to define a matching partition: this will enable you to control which data will be affected by mutate, and keep the file signature sound.

g = Genome("some/image.bmp", "output.bmp")

# apply the discussed partition system
g.partition([
    ('head', range(0, 0x36)),
    ('raw') # if no range is provided for the last one, it will cover up to EOF
])

g.mutate(.50, 0x7F, ['raw']) # only affect the 'raw' partition

You can also, if you have two similar file (in size and signature) set up arbitrary partition for the data before using crossover: the result will be a mashup of both file along the given partition (say e.g. partitioning line by line, with crosser function returning odds of self and evens of mate).