View source for Talk:Seraphim

Not that anyone cares to much. But I am working on rebuilding this robot.

The New Gun

I used the understanding of how her old segments were suppose to work, along with some old research, new research, and cross examination of many current open source robots to come up with a new set of all round segmentation for her gun. If you look closely you can see the correlation between the old Seraphims non adaptive scores and the new ones. However, better technology, understanding and some extensive tweaking, has allowed me to make gains against surfers (and also I think it is mostly how my new stat buffer choses the best bin, and how the bin scores are compiled).

Here are the scores for the gun up to my current tests, after it started to be somewhat competetive. These are the quick learning version of TC 2007.

Name	CC	RMX	SHA	WS	WOE	Surf	DM	FT	GG	RMC	WLO	No Surf	Total	Comment
Seraphim 2-SR2	73.52	83.01	70.19	83.76	79.68	78.03	84.39	76.87	79.59	77.01	78.07	79.19	78.61	30.0 seasons
Seraphim 2-WY3	72.06	82.88	68.73	85.96	77.41	77.41	82.64	78.03	80.04	76.32	77.34	78.88	78.14	30.0 seasons
Seraphim 2-WY2	72.20	82.56	68.70	84.26	78.47	77.24	84.62	77.08	80.39	77.32	79.19	79.72	78.48	30.0 seasons
Seraphim 2-WY1	71.32	80.73	70.75	84.52	79.18	77.30	84.46	78.52	82.57	78.56	79.32	80.68	78.99	30.0 seasons
Seraphim 2-WY0	70.21	85.19	69.86	85.21	79.62	78.02	83.69	76.50	78.96	76.29	77.67	78.62	78.32	30.0 seasons
Seraphim 2-PI1	74.00	83.57	66.66	85.72	77.90	77.57	84.49	77.01	81.15	75.95	79.67	79.65	78.61	30.0 seasons
Seraphim 2-PI0	71.41	83.19	68.96	84.93	79.75	77.65	83.99	76.85	80.18	75.87	79.58	79.29	78.47	30.0 seasons
Seraphim 2-RA1	70.48	80.64	68.69	81.70	77.56	75.81	86.03	76.85	76.40	77.01	78.91	79.04	77.43	30.0 seasons
Seraphim 2-RA0	71.45	82.14	67.41	83.29	77.38	76.33	86.38	74.97	79.71	76.63	78.97	79.33	77.83	30.0 seasons
Seraphim 2-LW2	70.75	80.58	69.46	81.86	77.77	76.08	84.85	76.49	78.62	75.54	76.88	78.47	77.28	15.0 seasons
Seraphim 2-LW1	71.68	82.81	67.21	83.34	77.51	76.51	83.97	74.15	77.57	75.51	77.95	77.83	77.17	15.0 seasons
Seraphim 2-LW0	71.87	81.41	67.22	82.95	75.52	75.80	83.73	78.36	78.48	74.56	78.89	78.80	77.30	15.0 seasons
Seraphim 2-LR3	70.54	82.26	66.49	81.87	76.47	75.53	85.12	77.62	78.45	75.42	78.16	78.95	77.24	11.4 seasons
Seraphim 2-LR2	71.63	81.27	67.30	83.60	76.54	76.07	85.63	77.20	79.45	76.96	77.83	79.41	77.74	30.0 seasons
Seraphim 2.0LR	70.82	81.60	68.14	80.20	77.64	75.68	86.51	78.32	79.04	75.55	79.36	79.76	77.72	15.0 seasons
Seraphim 2.0.1MC3	71.75	82.14	68.00	84.74	78.29	76.99	82.23	77.78	77.19	74.87	77.40	77.90	77.44	15.0 seasons
Seraphim 2.0.1MC2	71.48	81.56	64.18	80.63	75.94	74.76	82.79	72.96	78.00	76.36	76.06	77.23	75.99	15.0 seasons
Seraphim 2.OldSeg	63.64	78.03	65.01	78.95	76.89	72.50	82.65	72.86	74.23	72.23	74.65	75.32	73.91	4.3 seasons
Seraphim 2.0.1NA	52.73	64.45	54.88	56.17	52.61	56.17	88.50	79.60	78.43	76.02	77.17	79.95	68.06	9.2 seasons
far too embarressing to show you these between here
Old Seraphim 0.1.63	66.33	81.68	65.82	76.16	71.51	72.30	87.35	79.24	83.52	76.58	79.51	81.24	76.77	15.0 seasons

• 2-PI1 - Unexpected but not unwelcomed increase against CC.
• 2-PI0 - Darn it DM, Stop dropping in score!

The New Movement

Seraphims old movement was riddled with bugs, I have gone back and corrected it somewhat already, but the old version is far to difficult to maintain. Her original movement was based off basic surfer, I have no intention of doing that again. Of course such a basic template is useful, I think I know a better way to go about doing it now. I have many many ideas for the new surfing movement, but it will probably use a similar segments to what the gun has. After that I will tweak the buffers to deliver better scores in the MC2k6, and perhaps some work in MC2k7 as well. That new anti-surfer challenge looks like it might be fun as well.

Name	BotA	BotB	BotC	WSC	APMC	CC	FHT	Sha	CFC	Total	Comments
Old Seraphim 0.052	97.30	92.35	79.00	89.55	32.65	30.04	50.54	26.26	35.61	52.60	1 Season

White Whales

http://a.imagehost.org/0146/dookideafeat.png

It beat Dookious there for a bit. — Chase-san 00:33, 20 July 2010 (UTC)

Hey, don't pick on Dookious! He's old and very slow... :-) --Voidious 01:13, 20 July 2010 (UTC)

But he pretty much defined the 2100 club as its founder! — Chase-san 03:04, 20 July 2010 (UTC)

Hehe, well thanks. =) And just kidding, I'd be happy to beat him, too! --Voidious 03:08, 20 July 2010 (UTC)

Aaargh, in that case, I have a truckload of White Whales. The only one of the top-20 I can beat is the ancient SilverSurfer (#19) Luckily I can harras a lot of lower ranked bots ;-) --GrubbmGait 22:56, 22 July 2010 (UTC)

Looks like it jumped about 33 ranks to around 24ish. The fix wasn't exactly complicated either once all is said and done (finding them was something of a pain though).

Considering it was pretty much just a flattener, I'm not surprised it scored about the same as SandboxDT =) I mean seriously, check out the comparison: http://darkcanuck.net/rumble/RatingsCompare?game=roborumble&name=cs.s2.Seraphim%202.3.0&vs=pe.SandboxDT%203.02 Same APS, same PL, just slightly different problem bots.

I thought SandboxDT 3+ was an actual surfer. But your right they are very similar in performance.

I am just glad I finally dug Seraphim out of the rank 50 rut it was in. I only changed about 14 lines of code. The big bugs were only about 3 lines worth, but I wanted to keep the flattener, and fix a minor bug I noticed.

I think SandboxDT was a little bit of surfing with a lot of flattening. But it's strange how such tiny bugs make such a huge difference. I left Neuromancer on the shelf for a long time because for some reason the movement just wasn't working correctly, until I discovered a nested iterator was calling the parent's iterator .next() method instead of its own. Amazing what a difference that made :-p

SandboxDT has a kind of adaptive Random Movement, where it merely adjusts the probability of reversing direction.

That was 2.x, I think 3.x uses some form of precise prediction.

Yeah its nuts. The big bugs were, an off by one error, an off by one error (in the other direction), and the wrong Boolean value. I mean honestly.

The complexity of the bots amplify the effect of "minor" bugs. If I flip the sign when calculating danger in Wave Surfing, which is adding only 1 character "-", and do the same when calculating density in the targeting, the bots score and ranking will drop all the way to near last place.

I think that actually happened to a real rocket test once. It literally flew the wrong way because someone put a dash where it shouldn't have been.

Off by one in the logic, rather then the code. To otherwise large numbers. The off by ones were for fire times. The Boolean handled the flattener.

Just noticed, this update puts Seraphim into the gigarumble, with Pear getting kicked out.

I actually mentioned that in nomination thread. But I wasn't going to push it. Seraphim is still an old bot.

I say we just stick to top 30 PWIN, no nomination necessary. If anyone's interested, feel free to update the participants list and/or move it to an official subpage of RoboRumble/Participants. There were several changes from the current list IIRC.