Assembly-Language Script Evolution

Denis Ivanov · 2007-08-21T11:50:52

I'm making an evolver like Tierra based on an assembler-like scripting language similar to the "Little Man Computer" model ( http://en.wikipedia.org/wiki/Little_man_computer ). Here is the instruction set: 3xx - ADD - Take the value stored in mailbox xx and add it to whatever value is currently on the accumulator. 4xx - SUBTRACT - Take the value stored in mailbox xx and subtract it from whatever value is currently on the accumulator. 5xx - STORE - Take the value from the accumulator (non-destructive) and store it in mailbox xx (destructive). 6xx - LOAD - Take the value from mailbox xx (non-destructive) and enter it in the accumulator (destructive). 7xx - BRANCH (unconditional) - Reset the program counter to the value xx. That is, xx will be the next instruction executed. 8xx - BRANCH IF ZERO - If the accumulator contains the value 0, reset the program counter to the value xx. Otherwise, do nothing. 9xx - BRANCH IF POSITIVE - If the accumulator is 0 or positive, reset the program counter to the value xx. Otherwise, do nothing. 298 - INPUT - Fetch the value from the opened file's current line, put it in the accumulator (destructive), and move on to the next line. 299 - OUTPUT - Fetch the value from the accumulator (non-destructive), put it in the current line of the currently opened file, move on to the next line. 0xx - OPEN FILE - Open file number 0xx (can be any value from 000 to 099). This opens the file for INPUT or OUTPUT. 1xx - OPEN FILE - Also opens a file for INPUT or OUTPUT (can be any value from 100 to 199). 2xx - OPEN FILE - Also opens a file for INPUT or OUTPUT (can be any value from 200 up to 297.). In this case, "mailbox" refers to lines in the file that is running (not to be confused with the file is opened by the file that is running). The scripts will be in their own files with valid names numbering from 000 to 297. What are some good pointers for me? Something I'm wondering about is how will I be able to make sense of it once some mutations happen. I don't want to be looking at a file full of numbers, how do I help make it easier on my eyes to see if any interesting mutation developed. I'll have to make the very first self-replicating script by hand. It'll be a script that copies itself to the file with a number one higher than itself. EDIT: And mutations will happen randomly during OUTPUT by having the 3-digit decimel string converted to a binary stream and having one random digit inverted. So for example, when any time a script would OUTPUT a line to another file, there would be let's say a 1 in 1000 chance of a mutation. If this occured, the 3-digit decimal that would normally be outputted to this other file would be converted to a binary stream. Let's say if the decimal was 152, then the binary stream would be 10011000. A digit from this would be picked randomly and inverted, so let's say it picked the fourth digit 100[1]1000 and inverted it to 10001000, turning the decimal value into 136. EDIT: The script interpreter that runs the files will be a console window that will randomly pick a file number, execute that file, and once that script has finished, move on to another random file. But one problem I see is that the script interpreter might get stuck in an infinite loop and never be able to move on to other files (if for example a mutation developed where a BRANCH statement made the script repeat over and over again). How would I prevent this? [Edited by - polyfrag on August 19, 2007 6:31:30 AM]

Artificial Intelligence Programming

Started by polyfrag August 19, 2007 06:04 AM

10 comments, last by Vorpy 17 years, 6 months ago

polyfrag

Author

2,504

August 21, 2007 02:20 AM

Quote:
call routine by searching forward for the matching label

How would it search for the matching label? What would the instruction call look like?

Quote:
If the jump address is specified in the instruction itself then programs will have difficulty jumping into the code of other programs, or even just randomly combining in a somewhat meaningful way. Or maybe it would still work, I don't know.

Yeah, true, the search space would be too big.

What if I used a Scheme-like language for the scripts.

For example: (+ 2 (- 4 5))

Which in diagram form would be:

   +  / \ 2   -    / \   4   5

This would resolve the issue of having too big a search space because the program could only be recombined in meaningful ways. But I don't knot about how reproduction would be done. There would have to be some system that traverses the entire tree and copies each node? I might as well make a 'reproduce' command, but then the whole script would be pointless. I have to break it into steps that could be recombined in other ways to serve useful purposes...

[Edited by - polyfrag on August 21, 2007 2:20:00 AM]

Vorpy

869

August 21, 2007 11:50 AM

Linear assembly code is really no less meaningful than a tree. It is the instructions themselves that determine what can happen. Although a tree does offer some advantages for genetic algorithms, and I think it is used for that sort of thing sometimes.

Searching for templates is done by simply scanning the memory for the matching template. You can use whatever instruction name you want. Look at the instruction sets in Tierra and Avida to see what they did.

Assembly-Language Script Evolution

This topic is closed to new replies.

Popular Topics

Recommended Tutorials

Assembly-Language Script Evolution

This topic is closed to new replies.

Popular Topics

Recommended Tutorials

Reticulating splines