Re: Question about Council Room data

[Rob Renaud]

(+councilroom-dev, so this is archived and others can help you as well)

I don't admin the councilroom server anymore.  Mike McCallister does.

The logs on isotropic are going to stay up even after you can't play it anymore, and you can directly pull them from councilroom anyway, so you have two layers of redundancy and nothing is changing.

Doing dominion stats hacking/analysis is certainly fun and a good way to learn how to do data analysis.

But I am not sure what you are trying to measure.  What do you mean by " cards tend to be the most definitive of the kingdoms they appear in?"

Rob

On Tue, Mar 12, 2013 at 11:57 PM, Mike Decuir <mde...@gmail.com> wrote:

I just started tinkering around the data available from Council Room and was interested in digging deeper, but I don't think my python skills are quite up to snuff to be able to pull down a local repository of the game logs to play with before isotropic goes down.  This is particularly true because I currently do not have an easily accessible linux box to follow your recommended setup on.

Do you intend to keep your repository up after isotropic goes away or do I need to hustle to get a copy of this data before isotropic goes down?

I am looking at this as an opportunity to develop some chops with using python for data analysis and hopefully contribute something valuable to the dominion community.

One metric I am particularly interested in exploring which cards tend to be the most definitive of the kingdoms they appear in from a statistical perspective.

Any advice you could provide on how to get access to this data with isotropic going down soon would be greatly appreciated.

Thanks,

Mike Decuir

[Mike Decuir]

Sorry for the brevity of that, I was trying to finish up the email just before I left.

I am starting to question exactly what I meant by that phrase myself because as I think through my idea, it isn't working out as I first thought.

I wanted to apply the mutual information criterion to look at both win/loss rates with, but also win/loss rates without when the card is present in the kingdom.

Further, I just wanted to just see what other interesting ideas I could come up with when I started exploring the logs.

Knowing that the isotropic logs will stay up after the site goes down is nice to help me relax and take my time understanding how to setup councilroom locally.  Unless, of course, there is a way to pull down larger data sets from the various pages of councilroom than I have been able to.

Thanks,

Mike

[Rob Renaud]

I am still a little unclear on what you mean by win/loss rates when a card is present in the kingdom.  You really want to know if a card was gained/not gained by a particular player.  I don't think that mutual information is going to be any different than just the win rate with/without correlations, because winning is binary.

One interesting stat I'd like to see is on the card winningness by turn graphs pages (they aren't rendering properly now on Chrome at least, but they are here http://councilroom.com/win_rate_diff_accum.html?cards=Cost%3D%3D5%26%26Actions%3D%3D1), is a control for amount of money earned in a given turn.  You'll see an effect where most $3 cards tend to get worse in the mid game, but that is mostly explained by the idea that earning only $3 in the mid game is bad.  So you can get a bunch of money per turn vs win rate curves as a control, and then compare the winningness per turn when a card is gained vs those controls for a given card cost to get rid of the 'povery == bad' effect, and focus on the cards contribution beyond that.

[Mike Decuir]

I think I need to think through what I am after a bit more.  The information may already be contained in the win rate with metric but I am not certain of that.

I am interested in finding cards that have more wins with and more losses without wins without and losses with or the other way around.  I realize that is a convoluted way of saying what I am after, but I am struggling to articulate my thoughts.

I am going to think through this a bit more and hopefully can articulate my ideas better.

Mike