|
Simple tabling of talent win and pick rates
Winrate field is the relative win-rate when using that talent, compared to games skilling the alternative choice
5 Best talent choices
10 worst over-picked talent choices Most over-picked are over 50% pick-rate, yet have lower relative win-rate than the other choice. (Most under-picked not shown as it's basically same table with name and other choice swapped)
Choices player-base is 'most wrong' about multiplies the offset from 50% pick-rate with win-rate offset*
I would rather let people come to their own conclusions about the reasons behind the data, rather than me impose a viewpoint. If anyone wants to do an article explaining these results feel free to use I got the stats off dotabuff myself anyway
Some more in-depth json available here**: http://pastebin.com/i1bz0rL2
* Therefore if they have negative relative win-rate -2%, but they are also picked 47% of the time. Their combined value will be -2 x -3 = +6 indicating people generally make the right choice with this talent.
** really I should have just put all the talents in the json for people to use but woops!
|
Where did you get your data from ? I think your results are skewed due to the fact that people skill some talents when they are losing.
But to control for that I would need other data.
For example you could compare the winrate of talents of teams that are behind in networth at the X minute mark. You would at least eliminate all the stomps. I'm pretty sure when people are stomping they never pick the -respawn talent, and when you're stomping the chances are very high you'll win in the end regardless of talents. It's just an example.
My point is by just comparing winrate of talents you are ignoring that there are a lot of factors that decide what talents people pick, and those factors might have a way bigger impact on winrate that a single talent choice.
I don't mean to be disconcerting your post was interesting, but yeah control for other variables if you want to make authoritative statements "like this talent is better".
Regards
|
data is from https://www.dotabuff.com/heroes/shadow-fiend/builds sections of dotabuff
Nice ideas, I'll definitely take stomps/clowny games into account. I am planning on doing further stuff (what talents are good against what line-ups), but I need 'proper' data from dota2 API and my scraper hasn't got enough games downloaded yet.
Just did this in the mean-time whilst waiting to have enough of my own data.
(I would be surprised if other variables could cause that big of a swing in winrate on their own. But I'll definitely check when I can)
|
Generally speaking respawn timer seems like a bad talent in many cases, especially when compared to +MS which isn't valued as highly as it should be. The effect of dodging a gank by moving out of it or not is hard to judge in most cases while stuff like being in the right spot a bit faster on the farming/fighting side is easier to see when playing.
|
people actually pick the +2 wolves on lycan, huh?..
|
It seems fairly conclusive but also it's fairly straightforward that you pick -respawn in games where you are dieing (losing) and not pick it in games where you are less likely to die (winning). Seems hard to adjust for.
That said most analysis on initial talent release was that -respawn was not that great but it's gotten more and more popular because it "feels" so good in game. It's quite possible that feeling is wrong. Very interesting.
|
I agree with the "already winning/losing" bias, especially when it comes to respawn time, I think it plays a big factor in the shown result.
That being said I'm shocked with Nyx talent... which I just checked on gamepedia :
The damage bonus to Spiked Carapace increases the damage to 200%, not by 200%
So I guess it's not as huge as I thought. But still, surprised that people do not just go for that, and that when they do, it's apparently "worst" than the other choice...
|
Even beyond respawn rate there are other biases.
The Phoenix Level 25 Talents (Supernova +2 hits, Supernova +1s stun duration) has a winning/losing bias. The further ahead you are and the more successful your Supernovas have been the more attractive the +1s stun is, the longer it takes you to get 25 or the more of your eggs are being popped, a sign of losing, the more likely you will take the +2 hits.
Though it is interesting that some of the respawn rate talents are absent from the list (like Phoenix's).
|
Baltimore, USA22222 Posts
61% of people picking +100 damage on Omniknight instead of the +6 mana regen boggles my fucking mind.
|
Northern Ireland22201 Posts
On March 31 2017 18:32 EvilTeletubby wrote: 61% of people picking +100 damage on Omniknight instead of the +6 mana regen boggles my fucking mind. do u really need +6 mana regen by the time ur lvl 20? when presumably u have arcanes/greaves, lotus orb etc...? omni isnt really a spell spammer either
|
yeah so I think I need to be analysing 50% of games where talent is skill when 'ahead'. 50% when behind, and that shouldremove bias for skilling certain talents when losing.
(will probably rule out some 'outlier'/really funky games as well. maybe complete stomps shouldnt count, people get too clowny)
Ive got a few hundred thousand opendota api game results from latest patch on my laptop now. Fingers crossed the skilling has timings assosciated with them, (I know theres gold lead at different game stages already there)
|
Baltimore, USA22222 Posts
On March 31 2017 19:47 ahswtini wrote:Show nested quote +On March 31 2017 18:32 EvilTeletubby wrote: 61% of people picking +100 damage on Omniknight instead of the +6 mana regen boggles my fucking mind. do u really need +6 mana regen by the time ur lvl 20? when presumably u have arcanes/greaves, lotus orb etc...? omni isnt really a spell spammer either
I can honestly say Yes, I've still run out of mana even with Greaves at that level.
What does the +100 add? Very rare that I think that matters if you're a level 20 support Omni, as the carries will be at or close to 25 anyhow.
|
|
|
|