Page 2 of 4
Re: Experience stats from 250+ replays
Posted: 24 May 2011, 23:51
by BaNa
interesting comparisons can be made, imo this would be a good base for any balance changes if you want to make them. (for instance sam vs packo -> sam is obv doing much better)
Re: Experience stats from 250+ replays
Posted: 25 May 2011, 01:03
by Beherith
Here is the ranking, with nanoframes removed, and only units that actually died tallied (no units left alive after game end were counted in)
Remember 0.19 is break-even
Oddities are the core and arm t2 arty having such a high difference - guess that can be attributed to skill - i may have to rerun and log the ranks of players as well, possibly using only results from veterans.
Code: Select all
Catalyst 0.81183481
Silencer 0.402006402
LLT 0.383447258
Leviathan 0.368404736
LLT 0.354053666
Beamer 0.341532702
HLLT 0.329752323
Pillager 0.314540265
Shiva 0.304821512
Sharpshooter 0.301964109
Buzzsaw 0.295161776
Pit Bull 0.295067922
Gaat Gun 0.294730478
SAM 0.289476295
Sentinel 0.286486193
Defender 0.280817077
Pulverizer 0.279204832
Dragon's Maw 0.278700839
Goliath 0.278219912
Tsunami 0.275327992
Chainsaw 0.267237081
Serpent 0.262688945
Sentry 0.260494336
Rapier 0.259323384
Hurricane 0.257478997
Eradicator 0.256941063
Big Bertha 0.255150545
Doomsday Machine 0.247072756
Luger 0.241180509
Stinger 0.238970237
Fatboy 0.238068067
Sabre 0.236481387
Shredder 0.234009053
Cutlass 0.233704632
Viper 0.233292666
Millennium 0.232390236
Penetrator 0.232304176
Slasher 0.229006377
Guardian 0.227282851
Thunderbolt 0.225612533
Banisher 0.225320354
Samson 0.22424744
Toaster 0.224121851
Pack0 0.221601892
Punisher 0.220430086
Executioner 0.220119557
Storm 0.216355763
Pelican 0.216288451
Urchin 0.214377437
Annihilator 0.211373808
Leveler 0.208657532
Supporter 0.207486227
Juggernaut 0.206751482
Copperhead 0.205497212
Phalanx 0.205252303
Typhoon 0.202036641
Dragon's Claw 0.201696797
Sumo 0.201161389
Nixer 0.200792549
Shark 0.199904902
Poison Arrow 0.197926231
Intimidator 0.197292008
Skimmer 0.193960439
Piranha 0.193591331
Brawler 0.192137152
Decade 0.191724213
Conqueror 0.191533763
Phoenix 0.191279041
Harpoon 0.189284866
Thud 0.188801548
Janus 0.188210449
Enforcer 0.18803482
Screamer 0.187379082
Vulcan 0.186395824
Can 0.186124277
Maelstrom 0.186077609
Crasher 0.185609283
Dominator 0.18512695
Karganeth 0.183879781
Snake 0.183145382
Crusader 0.182214013
Recluse 0.180841915
Shellshocker 0.180732079
Rocko 0.180704688
Wolverine 0.179802902
Jellyfish 0.179296257
Retaliator 0.177546471
Lurker 0.177208501
Archangel 0.175665221
Krogoth 0.174305524
Peewee 0.173857291
Searcher 0.17325151
Jethro 0.173012826
Shadow 0.172709682
Flakker 0.171491058
Thunder 0.171046483
Warlord 0.170688145
Scrubber 0.169943986
Lamprey 0.169774053
Panther 0.168627677
Pyro 0.168065489
Lancet 0.167940337
Stumpy 0.166928285
A.K. 0.166708649
Morty 0.166672195
Catapult 0.166467524
Hawk 0.165516333
Raider 0.164451752
Skeeter 0.163903226
Wombat 0.161171711
Croc 0.160733067
Anemone 0.159890601
Halberd 0.157246255
Zeus 0.156015638
Behemoth 0.155262988
Anaconda 0.154977396
Detonator 0.154427603
Gimp 0.152841711
Vamp 0.152278564
Triton 0.151670061
Fido 0.14853258
Hammer 0.148518628
Messenger 0.148509629
Bulldog 0.147650535
Reaper 0.147643678
Snapper 0.147553927
Razorback 0.146359899
Ambusher 0.145488661
Vanguard 0.143013888
Flash 0.142323599
Merl 0.140453417
Exploiter 0.138557947
Stingray 0.138394128
Moho Exploiter 0.134493919
Avenger 0.134358565
Banshee 0.133139762
Instigator 0.130502653
Freedom Fighter 0.126619579
Warrior 0.125504901
Archer 0.122529475
Liche 0.120898148
Pincer 0.120283217
Slinger 0.119540434
Titan 0.1178728
Garpike 0.116959583
Swatter 0.116511238
Cobra 0.115953916
Maverick 0.11304778
Ranger 0.111702996
Zipper 0.104189689
Tremor 0.103003609
Gremlin 0.102629289
Krow 0.099645685
Voodoo 0.096236315
Mercury 0.095806522
Blade 0.095677868
Diplomat 0.093984143
Weasel 0.086645206
Manticore 0.08474442
Jeffy 0.084606012
Bantha 0.083155778
Moray 0.074818164
Flea 0.074649467
Marauder 0.074289315
Termite 0.074262165
Stiletto 0.062778452
Commander 0.061804483
Commander 0.057698652
Tornado 0.044688118
Bladewing 0.037205689
Spider 0.032825761
Commando 0.025596312
Dragonfly 0.023456459
11 0.013504911
Commander 0.012680355
Kilo 0.006468645
Micro 0.004858742
Mega 0.003386086
110 0.002635396
Commander 0.001417325
1100 0.001265162
Re: Experience stats from 250+ replays
Posted: 25 May 2011, 02:14
by Cheesecan
Boy imma make me some more of those 1100's.
Re: Experience stats from 250+ replays
Posted: 25 May 2011, 02:20
by Johannes
Cheesecan wrote:Boy imma make me some more of those 1100's.
You should, they're definitely some of the most costefficient, underused units in the game. Even if their kills don't show here properly cause they were dead then.
Re: Experience stats from 250+ replays
Posted: 25 May 2011, 02:26
by triton
Eh I'd like to know how many 1vs1s you have on these stats.
Re: Experience stats from 250+ replays
Posted: 25 May 2011, 08:38
by Beherith
All team games on random maps, not enough 1v1's to do this.
Re: Experience stats from 250+ replays
Posted: 25 May 2011, 19:14
by knorke
Some units are more likely to rake in "free" XP when they destroy bases in games that are basically already over.
For example if the defense are already breached and a stream of pyros gets spammed into the base then the pyros XP will go up even though any other unit could have been used. In that case it does not tell much about the combat value and the pyro would be rated too high. On the other hand some units might be rated too low ie AA turrets.
To get stats that really represent combat value, one would have to end recording as soon as the game is "won" as not to include the mobbing up into the stats.
Re: Experience stats from 250+ replays
Posted: 25 May 2011, 19:26
by Beherith
Knorke, it works that way: an unit xp is only recorded now if the unit dies. If the unit is doing free endgame base rape, it probably wont die.
Re: Experience stats from 250+ replays
Posted: 25 May 2011, 19:34
by TheFatController
I suppose Commando doesn't get the damage from its minelaying attributed to it (or it fails a lot, probably both!)
Re: Experience stats from 250+ replays
Posted: 25 May 2011, 19:34
by Pxtl
In that case, how the hell is the Buzzsaw so high? I can't imagine a buzzsaw destroying a substantial amount of terrain and *not* ending the game.
Re: Experience stats from 250+ replays
Posted: 25 May 2011, 20:24
by knorke
There are only 2 buzzsaws that got high XP:
http://beherith.eat-peet.net/stuff/unitexp/Buzzsaw.png
not enough to conclude anything.
So if xp is only recorded if the unit dies, then some units are maybe rated too low?
ie turrets like the Mercury: (thats the longe range AA missile isnt it? never remember the names)
http://beherith.eat-peet.net/stuff/unitexp/Mercury.png
Mercury 0.095806522 with new stats
They might contribute alot to the victory of their team but the most effective ones (the ones that kill alot and do not die) are excluded from stats.
Its tricky no matter how you turn it
Re: Experience stats from 250+ replays
Posted: 25 May 2011, 20:55
by Pxtl
Exempting immobile units from the "must die before endgame" would help.
Re: Experience stats from 250+ replays
Posted: 25 May 2011, 20:58
by smoth
There has to be a way to tap all units and record their stats at game end
Re: Experience stats from 250+ replays
Posted: 25 May 2011, 22:39
by Beherith
Smoth, there is, and this script does that as well.
I did use those stats at first, but it doesnt make sense to count the XP of a unit that didnt fight - while a unit that died probably fought to its highest extent.
Re: Experience stats from 250+ replays
Posted: 26 May 2011, 00:27
by albator
I just would like to point out that these start are really nice to know about 1v1 when unit are fighting. However, the fact that the unit can be spam fastly or are really fast and able to raid key stucture is not embedded in thoses stats. That the reason why phoenix does not have so nice xp even it is supposed to be op cause it destroy key stucture. This is another story for Hurrican but I think that is biased becasue of micro of the first 5 t2 bomber spam on 8v8 maps for which poeple choose core bombers cause of their hight dps and micro them like hell.
This was just to point out that raiding unit that have a high xp are truly good unit. And panther has better xp than stumpy which is supposed to be already realy good. That the reason why I still think it deserves a +30% E cost.
About croc, i think you should pay for the amphibious ability and that the xp should be quite low for that kind of unit. That the reason why I still think it deserve a +30% E cost.
It would be nice to get the stats on spads2 (2v2 3v3) with a filter on the game (like the best 20% of games whose damage dealt per player is the highest after 15 minutes) so only game where good player are involved are taken into account. i think it is better than taking game with vet cause of smurfs and sucky vet.
Re: Experience stats from 250+ replays
Posted: 01 Jun 2011, 01:36
by Beherith
I would like to request a few hundred DSD replays, if any autohost owners have them, I would like to run them through this stats collector.
Re: Experience stats from 250+ replays
Posted: 06 Jun 2011, 20:13
by Forboding Angel
No offense behe, but this sort of stat collecting is useless. The reason why is because of the way that spring calculates experience. A unit that fires 50 times a second and does a grand total of 1 dps to a larger unit, will have experience in the millions in just a few seconds.
Hopefully this illustrates the point.
That's why the pyro accumulates experience like crazy. Same for flashtanks and other rapid fire units.
The experience system in spring will have to be fixed before any experience based stat collecting can be useful.
Re: Experience stats from 250+ replays
Posted: 06 Jun 2011, 20:17
by smoth
I'd like to see you actual data on that forb. 1mil xp is a huge huge number. (been doing actual work to read xp lately)
Re: Experience stats from 250+ replays
Posted: 06 Jun 2011, 20:55
by Beherith
I'm pretty sure you are wrong forb. Flash and pw would gain way more xp than ak and gator.
By the way, please note that experience gained is proportional to:
(damage dealt to enemy / enemy units health) * (enemy unit cost / own cost)
Pyro gets so much XP cause it has about 3 seconds to bring back its cost before it dies in battle. And because it has INSANE dps.
Ergo, this is the single best way of collecting a units value wrt other units. Prove me wrong with a constructive proof.
QED.
Re: Experience stats from 250+ replays
Posted: 06 Jun 2011, 20:56
by knorke
A unit that fires 50 times a second and does a grand total of 1 dps to a larger unit, will have experience in the millions in just a few seconds.
Hopefully this illustrates the point.
That's why the pyro accumulates experience like crazy. Same for flashtanks and other rapid fire units.
From looking at rts\sim\units\unit.cpp it does not look like that.
Normal damage dealing:
Code: Select all
if (damage > 0.0f) {
recentDamage += damage;
if ((attacker != NULL) && !teamHandler->Ally(allyteam, attacker->allyteam)) {
attacker->AddExperience(0.1f * experienceMod
* (power / attacker->power)
* (damage + std::min(0.0f, health)) / maxHealth);
damage is a float and there seems no rounding/clamping to integer to be going on.
This looks like the unit who gets the actual kill gets a "bonus":
Code: Select all
if (health <= 0.0f) {
KillUnit(false, false, attacker);
if (isDead && (attacker != 0) &&
!teamHandler->Ally(allyteam, attacker->allyteam) && !beingBuilt) {
attacker->AddExperience(expMultiplier * 0.1f * (power / attacker->power));
teamHandler->Team(attacker->team)->currentStats->unitsKilled++;
Cant see why this would mean "fast firing = more xp" either.
Now I didnt look at this in detail but imo
