|
|
Darwersi History
Staring:
- Yves Lafon (AB/IOS guru)
- Stéphane Lavirotte (GUI/Windows master)
- Olivier (me) (all sorts of useless things
 )
1999-
- Oct 14: Some more weeks later... news on freshmeat.
- Nov 14: Some weeks later... added to freshmeat.
- Nov 8: Added 1.4M disks for windows install.
- Apr 14: The new version is available on INRIA's ftp. We have to wait some weeks for the beta testers reports before a more official announce.
- Apr 10: Ready to release a slightly improved 2.1 version... but this time it works under windows!
- Apr 5: Well, the 2.1 release is under its way. Stef has made some nice improvments in the GUI. (He is now using the BWidget toolkit.)
- Mar 28: We switched from Borland C++ to Visual C++ and... Darw has recovered its mind! No doubt it was a compiler problem. Thanx to Pierre for his help and Stef for his courage. Damn'! All that time wasted for such a stupid bug!
- Mar 25: After 3 days of install/reboot/uninstall/reboot/install/crash I managed to use boundchecker... and it detected absolutly no glitch in Darw's code. (BTW: I was very confident with that because Purify was very happy with Darw running on unix...). No idea for the next step.
- Mar 20: Seem's to be a long time since last log! I've worked on other (non darwersian) topics and the darwersi's windows bug is still on my TODO list. Fixed some minor problems thought...
- Jan 15: This bug is really a mystery... and I have to use my spare time on another project. Sorry for all of you that are asking for a patch.
- Jan 12: Ok, Ok... the windows version is buggy... it seems to be a nasty one. Darw is playing like a real braindead. Linux version is fine... of course! It's really a pity if we consider the download counters. More than 500 downloads for windows version and 120 for unix within the last 15 days (with almost no public announce)...
1998-
- Dec 23: Done... We decided to put it on the INRIA server for download but we made few advertising... wait and see.
- Dec 20: Release is almost done... Xmass is our target.
- Nov 1: We are beta testing the release package...
- Oct 24-25: Damn' Darw made only 2.5 wins. No more doubt, D has been badly hit by stronger programs at PRII. No MPC and no big opening book were the main reasons for this failure. See you all at PRIII (aka D's strikes back).
- Oct 23: Nothing really new... Yves seems to have slept a lot. We're thinking of a new end search ordering scheme based on our gigantic end pattern base. We have to find out how to invert a one billion term matrix.
- Oct 20: Yves took two days of vacations for a so called 48H hack... I'm curious to see the kludge he's able to add that way .
- Oct 19: Packaging all this mess for release is not that easy. Wrote some more (or updated) HTML help.
- Oct 18: I've found and coded a way to normalize these bumpy evaluation factors. Darw was suffering fast and drastic evaluation drops. This seem's to be fixed. A really good opening book is now under its way. Alas there is few hope to fill it enough before PRII. This is also a good step forward MPC.
- Oct 16: 2300+ IOS (8th).
- Oct 15: I've done some little speed improvments on evaluation function code. Pattern access is now working fine even under harsh memory conditions (dynamic load with predictive caching).
- Oct 15: Yves is working on an ES-like transposition table for midgame. This sounds cool but time is running fast these days.
- Oct 14: We are seriously planning to release Darw during the famous PrincetonII computer championship. Only one week before deadline... Could we do it? (I mean, could we do it and survive one full night-hacking-week?)
- Oct 5: Urgh... I'm back and the evolution I've carefuly started before leaving has been aborted to fix one of the broken cluster nodes. This is a hard hit... The new brain I was tuning won't be ready for PrincetonII and I have to exhume the old primitive amoeba's brain.
- Sep 5: Damned I have to left this mess for vacations.
- Sep 3: Yves reworked a lot his EndSearch algorithm... WLD at -25 and exact at -24 has been sucessfuly tried in real conditions.
- Aug 30: Something is obviously wrong... but where? Switching back to tje old patterns.
- Aug 10: Major rework of my patterns deductive builder (one more) is done... That was a tough piece..
- Jun 15: Why is that braindead new Darw so corner greedy? Added some kludge to pattern factory to prevent excessive correlation of patterns.
- Jun 11-14: Another rework of POV scripts... This one is the last (until the next one of course).
- Jun 10: A new big-bang occured this day at 15:00PM. We all know this could take some time to produce a valuable specimen. But this time I'm using far much refined selection scheme and the distribution method is more efficient. The virtual cluster is composed of a nice mix of 11 Dec(333->500Mhz 384->512Mb) 3 Sun (Ultra 10/30 166Mhz 256Mb) and 25 Linux boxes (plenty of (dual)PPro200 and some PII333 ranging from 128 to 256Mb).
- Jun 6: 2200 IOS points... 800 to go.
- Jun 5: A special end-search pattern collector is born. This should provide Yves with a far much accurate ordering in D's end-search alpha-beta.
- Jun 1: Spread the 2.0a1 version to some friends (mainly windows beta testers).
- May: Yves is working hard on end search. He's restarting from scratch with a new move generator. JCW's code has been patched far beyond hope and is 'cvs removed' with extreme pleasure.
- May: Build the new brain factory. Darwinian selection is now using a pyramidal selection scheme (something like pools used in sport tournaments). This time I added a retro control mehod that should ease the evolution interpretation for these sub-god individuals that have to tune it ;)
- Apr: Thrown away the old distribution toolkit. The new one is far much robust and flexible.
- Mar: Yves added a transposition table to end-search. At least we are able to solve -21 in tournament conditions.
- Feb: Reworked the pattern gathering/deducing mecanism. It is now fully generic and should allow me to experiment with a wider range of patterns. The one with 64 squares seems to be promising.
- Jan: 2100 IOS points... need 900 more ;).
1997-
- Dec: At least we broke the 2000 IOS points wall. Nights a really too short for my TODO list.
- Oct 18-19: First tournament for this newbee Darw'. PaderbornIV is very exciting and D managed to reach the impressive rank of 14th (hum... there was only 15 players)!. Useless to say that Darw was something like a sweet jawbreaker for the big beasts here (Logistello, Hannibal, Turtle, Spock). At least honour is safe with some nice wins and a very surprising 37/27 against Brutus.
- Sep: Yves added IOS protocol to Darw allowing it to achieve some nice loss ;)
- Aug 4-7: Takeshi Murakami is defeated 6-0 by Logistello... Darw is coming too late ;).
(rating oscillates between 1700 and 1800 IOS).
- Jun 13: Yesterday Stef made the first compilation of the Darversi2 library on Windows. The GUI is growing fast and I have to rework all the POV scripts to produce nicer pictures for the greedy Stef.
- May: A new monomaniac is joining the team: Yves Lafon. His first aim is to improve the dumb AlphaBeta that is reaching painfuly 7-8 plies.
- Mar 10: A friend of mine, Stef, is looking at me struggling with a (very evolved) bug and suddenly he says: 'And what about an unified Unix/Windows interface with the Tcl/Tk?'. I guess he had smoked something unusual but it was too late to recover from this BIG error.
- Feb: Something has to be done to fix some nasty bugs and a very 'human' touch that D is showing in all its loss. I've thrown all this old ugly code and I'm starting a completly new library that should reach a degree of perfection still unseen on earth.
- Jan: The first public release of Darwersi(1.1). Many thanks to Pierre and Patrick who wasted some of their precious working time to give Darw a windows GUI.
1996-
- Dec 15: End of the first version of the Darwersi's Artificial Intelligence. First tries are encouraging enough to help me tracking the creeping bugs and building a first GUI. Nice Xmass gift for an innocent father.
- Nov 16: A windy winter day in an hospital chamber, I'm asking myself what I could do to make world sunny and humans happy. And THE great idea strikes me like lightning. I will do one more of these useless othello programs. (The only reason why this obviously stupid idea didn't sunk in my wavy unconscious is that my ego awake at that point saying something like: 'Do you remember the many many othello games you lose against your dad? Well it is time to strike back!'. Since, I'm trying to convince myself that a so brilliant idea was in fact a side effects of drugs they were using on me...)
|
