MEP SPEL compatability

9 views
Skip to first unread message

Logan Streondj

unread,
May 17, 2016, 12:53:49 AM5/17/16
to me...@googlegroups.com
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi MEP and GP experts,

Intro:

I'm working on SPEL (speakable programming for every language) which
is programming language for making GI-OS (General Intelligence
Operating System). Anyways so now that I have the vocabulary and
syntax, I need standard libraries and things like that. But I've no
time to do that, and don't have anyone to help me, so last week I
discovered Genetic Programming which can synthesize programs, to
generate the aforementioned libraries.

Just an additional motivation to go the GP program synthesis route, is
that fault-tolerance is often based on redundancy, so it should be
fairly trivial to evolve multiple hero algorithms for various local
maximums, (that way can run 2 or 3 different algorithms at a time for
fault-tolerant safe-modes).

MEP questions:

Currently in the virtual machine, I'm using 32byte blocks, which fit
up to 15 instructions, averaging probably 4. I'm wondering if these
are similar to a MEP "chromosome", or gene.
The idea behind these 32byte blocks is that they each hold a sentence,
for example there can be 3 loads and a command in a sentence.

Additionally the encoding supports showing dependency, so a long
sentence could be multiple blocks, (if loading large constants).

Anyways so my main question is in relation to how SPEL can be MEP
compatible.

Thanks,
- --
Logan Streondj,
A dream of Gaia's future.
http://joyfullifestyle.ca

Speakable Programming for Every Language:
http://wyn.bot.nu/spel/src/vocab/gen/start.html


You can use encrypted email with me,
how to: https://emailselfdefense.fsf.org/en/
key fingerprint:
BD7E 6E2A E625 6D47 F7ED 30EC 86D8 FC7C FAD7 2729
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)

iQIcBAEBAgAGBQJXOdCdAAoJEIbY/Hz61ycpIdEQAJ96efmnja976cVRxcde6K7o
YJtjhwGiEW4/FjEFKaec6pAIUBsEEhrxo/zgGaknHiTbhoEpOVAgAb1WKFf/psDr
QJe8JrEFyYHCaQf5zahYhU9qx6l2LeErWKKSAVfxjqnibeU6bcsiCtvcrXrBwS5X
KBTtvPogyt57a2FEkGdNEvXR19o2sQGze+xXh9zaMX97t1WgFd3AnnnJu4U4EHDX
lwnEXXayLgBjVtOnePMLlYAfsNDmZvHHtgjzhXGW0/3kcjoGXDn6L6RlbG955EFz
HC6WGFmwD5OtpPwv4eLbFDu30hvjilWNs/aNK5gCTzR2LWGgHbjo5XngAjWtebPO
lb+NY/zTBcK+yP7JJRvbscmHl8GxAtfQOcfwQxc4vBL9OGA3vXFpHqInnlSCBlEd
4QohsNPhfByREnIyGRIWDcN/uHLk/Bb4ExCUjFcAbxOOs0i0FnHv1t6PB4EGO+7J
brE2soZeDEvYzZaZZTPsmuzGL2tEqhwucgAAaJLebv/jiYDUivjTCTkXKXExRZA5
ofH6OLCd3kWicK6S29XBDT0qBviCWtNBvoMlCpDk13Qsexh2JhWhWCHaWv9v9MJm
wUNh5YAfov4iiSR75M0E/r4OTiLENWixCcxT3vlQOjHqo6TySOTBQBIf0mgcubyH
1uackXfCPLf5ddnwm9Vx
=U1sP
-----END PGP SIGNATURE-----

Mihai Oltean

unread,
May 17, 2016, 2:09:01 AM5/17/16
to Multi Expression Programming
Dear Logan,

Thanks for showing interest in Genetic Programming / MEP.

Genetic Programming indeed is used to generate computer programs, but you should not expect to generate very long programs / libraries. In some cases even a 10 lines program can be difficult to generate because the search space (where we search for a solution) is extremely large. Just imagine that you have 20 instructions available and the program has 10 lines. You can form 20^10 possible programs (most of them of course are buggy), and you have to discover only one. It is like searching the needle in the haystack...

The challenge is to find problems for which humans don't have a good / optimal solution and to try to discover one or a better one.

Regarding MEP structure: it is nothing else but a simplified computer program, a sequence of instructions:

dest1 = i1 + i2;
dest2 = dest1 * 5;
dest3 = i2 < 4 ? i3:dest1;
....

What is different in MEP compared to other Genetic Programming variants is the fact that I can stop anywhere in the instruction sequence and return the best computer program in that sequence (Cartesian GP usually evolves what instruction will provide the output).

regards,
mihai
P.S. I had a look to your project and looks very interesting. I wish we have something like that in the future...
Reply all
Reply to author
Forward
0 new messages