EPT![[image loading]](http://img340.imageshack.us/img340/4124/examples.png)
It is hosted at http://userscripts.org/scripts/show/65257
Discuss/show your love/appreciation here. oberon will also be happy to answer any questions.
TLOdds - Odds information on TL posts
| Forum Index > BW General |
|
pangshai
Chinatown5333 Posts
It is hosted at http://userscripts.org/scripts/show/65257 Discuss/show your love/appreciation here. oberon will also be happy to answer any questions. | ||
|
jalstar
United States8198 Posts
| ||
Empyrean
17015 Posts
| ||
|
jalstar
United States8198 Posts
 Shine vs  Bisu on Fighting SpiritShine vs Bisu on Match Pointdoesn't take the map into consideration | ||
|
bluemanrocks
United States304 Posts
EDIT: doh! you just scroll over it..... what a doofus i am! | ||
|
ghermination
United States2851 Posts
| ||
|
Empyrean
17015 Posts
On December 30 2009 15:43 ghermination wrote: I really like this, and i'm going to use it, but i wonder if you could add a "map" function that found the win percentages on the map. Actually, this would somewhat address the problem of weighting by time. | ||
|
ghostWriter
United States3302 Posts
| ||
cgrinker
United States3824 Posts
| ||
|
oberon
United States1320 Posts
All odds are calculated based on lifetime record. ELO is not considered. Opponent strength is not considered. How recent games are is not considered. Player record on individual maps is not considered. None of these are bad ideas -- it's just that I'm getting all the data by parsing TLPD HTML, which is a pain. ELO should be relatively easy to pick out. If I have the time/energy to parse the full game log, instead of just the front page win %s, I could do recentness. Opponent strength is trickier. Player record on individual maps is (probably) worthless -- the sample size is tiny. Going 4-1 on a map is a lot less informative than going 70-30 vs. Zergs over a career. The next version will have a menu option that allows you to choose between BO1/3/5/7. I'll replace the "Win Chance" text with what the current mode is (e.g. "Best of 3"). The method for calculating the odds is already in the plugin -- just need to hook it up to the UI. Other feature requests welcome. --oberon | ||
|
MasterReY
Germany2708 Posts
| ||
|
oberon
United States1320 Posts
On December 31 2009 00:29 MasterReY wrote: Imho you should consider the overall racial stats on a map and not just the player stats on a map (like you mentioned: 4-1 on a map). Totally agree. As I mentioned, I wanna see how other people combine home field advantage info with log5 predictions, since HFA is a known percentage advantage (just like map racial advantage). That's what I was getting at in my first paragraph -- sorry if it was unclear. Unfortunately, I had my wisdom teeth out a few hours ago. It's really hard to code (or focus on anything, really) when leaning forward causes blood to dribble out of one side of your mouth. I'm hoping to shake that off and get the next version of this up this weekend. --oberon | ||
Heyoka
Katowice25012 Posts
| ||
Kinky
United States4126 Posts
 Though I'm curious as to what formula it uses. Is it the one that was posted a while ago using the player's last 10 and historical stats? And what's the difference between generic and racial? | ||
|
oberon
United States1320 Posts
--oberon | ||
|
oberon
United States1320 Posts
--oberon | ||
|
oberon
United States1320 Posts
--oberon | ||
|
Un Gato
United States123 Posts
: P | ||
|
oberon
United States1320 Posts
On January 02 2010 19:27 Un Gato wrote: Programming this with blood dripping from your mouth? Damn, you are hardcore. Useful, but in a slightly crazy way. : P Actually, things just healed a lot faster than expected...but I like your version better. --oberon | ||
|
hyst.eric.al
United States2332 Posts
too bad i use chrome  | ||
|
seRapH
United States9776 Posts
like in the cj/stars thread, u can find that the zero vs iris gives iris a 55% winning rate. which is absolute bs if you take their recent games into account. zero loses to FLASH, iris loses to saint/hyvaa | ||
|
oberon
United States1320 Posts
On January 03 2010 02:09 majesty.k)seRapH wrote: this really shouldn't be using lifetime records imo like in the cj/stars thread, u can find that the zero vs iris gives iris a 55% winning rate. which is absolute bs if you take their recent games into account. zero loses to FLASH, iris loses to saint/hyvaa Counting every game someone has ever played equally is an (obvious) weakness, but I was just trying to get SOMETHING working. The problem is that the TLPD page for players lists only their most recent 10 games for each matchup, which is too small a sample size. Even then, that doesn't include opponent adjustment. ELO helps somewhat with all of this, but it's not a perfect answer, either. To some extent, it's hard to provide really good predictive stats while working outside the server, as GM scripts must. Even given all this, however, if you look at results threads (where you can quickly compare predictions with actual results), you'll find that the odds are usually pretty accurate. I'm not saying the numbers are perfect, but they're definitely not worthless. --oberon | ||
|
Empyrean
17015 Posts
| ||
|
oberon
United States1320 Posts
-Fixed a bug wherein the wrong racial matchup would be used from map data, making all map-based numbers wrong. -Added ELO support The ELO support was a bit of a nightmare, but it's certainly a more sound measure of ability than simply historical win/loss record (though the two strongly correlate). I may need to find some way to cut down on the table size soon -- 5 rows isn't small. --oberon | ||
| ||
