[Home]David Alves/PhoenixArchivedChat

Robo Home | David Alves | Changes | Preferences | AllPages

Congrats David! Welcome to The 2K Club! -- Axe

Thanks Axe. :-) --David Alves

I couldnt believe you finally released Phoenix after 3(?) years. Congrats man. --deathcon

Having progress with your gun I see. Cool! -- PEZ

Actually my previous high mark in the normal TargetingChallenge was 91.99, so 92.08 is probably no better. But my previous top TargetingChallenge/ResultsFastLearning score was 88.76 (10th) and my new one is 89.63 (5th). :-) --David Alves

Awesome WaveSurfingChallenge2K6 scores! Seems 0.30.1 gives a new dimension to 'Crushing the weak'. Do you also have the scores for the rest of the MovementChallenge2K6? -- GrubbmGait

Yes, I am highly, highly impressed with those scores. You have replaced Corbos's Chalk as the tank I am very jealous of in the WaveSurfingChallenge. Nice job! -- Voidious

Here are some tips to help you improve your WaveSurfingChallenge scores:

That first tip seems like a good one; Komarious actually runs away whenever there's no bullet in the air, which is more of a codesize-restriction style tactic, but now you really got me thinking about it. I am skeptical of the value of anti-cornering code in real battles, but it has been a while since I removed it from Dookious, so maybe I'll give it another try... When you say "figure out what really matters", are you talking DynamicSegmentation?, or just EmpiricResearch when designing your tank? =) -- Voidious

I meant research, never tried DynamicSegmentation?... not successfully anyway. =) --David Alves

So Voidious asked me to run a 500 round TC against Shadow. Here are the results. Can anyone do better than this? (69.92 average bullet damage / round)

1st: davidalves.Phoenix 0.411	47280	7700	1540	34960	3080	0	0	154	346	0
2nd: abc.Shadow 3.66d	20760	17300	3460	0	0	0	0	346	154	0
--David Alves

"Very different surfing" . But i am curious what you use in melee because it is very succesfull in melee as well!!! It reminds me of the great Duelist*Melee range of bots. And do you use a different gun or segmentation in melee or is it the same as in 1-vs-1? --Loki

The melee movement is a minimum risk movement similar to the one in DuelistMiniMelee. The gun question is a little more complicated. While there is more than one enemy bot alive on the field, it uses a gun designed for melee. If Phoenix is one of the last two bots alive in a melee battle, it switches to its 1-v-1 gun for that fight. --David Alves

Holy *cow* dude, 20+ points to 2079! Awesome! That's easily the biggest jump I've ever seen in the thin air of 2050+. I guess the race to 2100 is on... =) Big congrats, dude. -- Voidious

Thanks. =) --David Alves

Yes, big congrats from Sweden too! That's an amazing score and it's bloody wonderful to see one of my mentors picking up the fight in the top race. Yay! -- PEZ

And congrats on officially nabbing the #2 spot. Best of luck to you in your quest for #1! -- Voidious

Actually data saving was introduced in 0.62. 0.63 was a lot of little changes to movement as well as better data saving. But I'm lazy when I type up the version histories. =P --David Alves

Version history for 0.6x is now updated.

It also presents a unique situation where Phoenix's rating going up might help Dookious to crack 2100. =) Go Phoenix! Seriously, though, great job on this PL monster of a version update. It could even end up undefeated if it gets a couple key rematches. -- Voidious

Umm, I feel a bit dirty for ganking your pre-loaded bot list - please don't hate me for it. :P -- Voidious

You've been firing 1.99 all this while?? -- PEZ

Yeah. Switching to 1.90 doesn't seem to have gained me any points, although there were other changes that might be affecting my rating. --David Alves

I found a magic number a while ago after creating a large spreadsheet with various calculations that seemed to make sense, mostly factoring in return on power, rate of fire, and escape angle based on the bullet velocity. My number wasn't 1.9x though. Once I started using it exclusively my bot sped way up and my rating jumped as well. -- Martin / Ugluk

Something like this? looks like somewhere around 2.4 is the best, in theory.

power 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
speed 15.8 15.2 14.6 14 13.4 12.8 12.2 11.6 11
max escape angle 0.530922556 0.554261834 0.579905879 0.608245579 0.639774963 0.675131533 0.715159313 0.761012754 0.814339942
p(hit) proportional to 1/escape angle 1.883513873 1.80420144 1.724417766 1.64407278 1.563049599 1.481192851 1.398289837 1.314038424 1.227988397
damage 6.4 7.6 8.8 10 11.2 12.4 13.6 14.8 16
energy gain 4.2 4.8 5.4 6 6.6 7.2 7.8 8.4 9
ticks between shots 13 14 14 14 15 15 16 16 16
goodness ((gain + damage) * p(hit) - power)/ ticks between shots 1.428095927 1.483721276 1.620480877 1.736083177 1.708152191 1.775425325 1.707712657 1.730355714 1.73123187

--David Alves

Something like that. My number is 2.5 .. beyond which is another tick of delay due to gun heat. In the four months of not looking at the code I've forgotten much of it, but I imagine I make exceptions for targets who are really close. -- Martin (Edit: 2.5 comes to approx 1.966 in your fomula)

I can't bust out the appropriate math off the top of my head, but it seems like you don't account for the fact that you are losing more power every time you fire a higher power bullet, a factor which would be multiplied by (1 - p), it seems. Speaking realistically, you'll rarely have a hit-% over 20%, so at least 4/5 times you are just losing that energy straight away. Just a thought, I am very interested in looking into this, but I have some faith in ~1.9 at the moment. -- Voidious

You're right. Updated the table. Same result though. The bottom number should be proportional to the "energy advantage per tick" of firing with that power, where energy advantage is the your energy gain plus their energy loss.--David Alves

I tested 2.5 (actually 2.49999) vs. 1.9 as a default bullet power. My rating dropped 2 points with 1.9. Not very drammatic, but I'll stick with 2.5. -- Martin

Hmm, thanks for the real world test results - in the end, that's really all that matters. I was thinking about this more yesterday... Wouldn't the "optimal bullet power" depend on hit percentage? Taking it to extremes, with 100% accuracy, a high bullet power is good; with 0% accuracy, low bullet power is good. Maybe this is obvious and is just something you consciously overlook when trying to come up with a "magic" bullet power. -- Voidious

Old versions of WaveSerpent did dynamic bullet power calculations similar to those above except p(hit) was hit percent/max escape angle. 1/max escape angle will give results more similar to 100% hit rate. With hit percent into consideration, 2.0 bullet power is the best power between about 8.5% and 20.5% hit rate. Of course, this doesn't take into account the value of high bullet damage for score, it merely shows the most energy efficient. -- Kev

Optimal bullet power depends on hit percentage? Ummm, well the problem with that is that your hit percentage is highly influenced by the bullet power used. Nice table up there, David, but it must have some errors. Look how it indicates that firing power 2 and 3 should be about equally good. Yet it isn't. I'll definately try 2.4 though. -- PEZ

Well, that's true, but my point is this: it seems to me that using high bullet powers is very bad if you are never hitting the enemy, but it's obviously very good if you are hitting them a lot. I still like your simple analysis of damage vs escape angle, and I'm still using 1.9 at present. But speaking of a theoretical "optimum" bullet power, it seems like it depends on how accurate your targeting is in the first place vs a given enemy. But hey, I've had a few drinks tonight, so I'll leave it at that for now... =) -- Voidious

Well, there's more to it than the damage/escape_angle ratio. If it was as simple as that then 1.72 something would be optimal. I remain convinced that the other influences, whatever they are, are minor in comparison. From my experience using 1.8 <-> 2.2 has never made much difference. But then I haven't experimented much with >2.2 and that might have been a mistake. Easy to correct though. =) -- PEZ

WaveSerpent used to calculated hit percent using (hit percent * average shot max escape angle / tested power max escape angle). This hopefully avoids the problem of hit percent based on bullet power. Vyper at one point used bullet power 2.4, but gained a few points by switching to 1.9, so I'm still pretty sure 1.9 is the best. I'm also sure staying close to a multiple of 0.5 is good because it improves your rate of fire. -- Kev

Best Phoenix yet in 0.81, right? What are the recent changes? -- Voidious

I'll update the version history tonight. --David Alves

Sweet, 2096 with 0.84... I bet you can taste it... =) -- Voidious

Had to do *something* to stop your PL rampage. ;-) --David Alves

Robo Home | David Alves | Changes | Preferences | AllPages
Edit text of this page | View other revisions
Last edited November 22, 2006 3:04 EST by David Alves (diff)