Drafting

[juzamjedi]

Pure rating on avg human draft picks, on color, including artifacts. Sounds like a huge improvement, let's do it! I look forward to seeing the AI draft smarter decks.

[delirimouse]

I was wondering, with Dark Ascension just around the corner, if you had given thought to what the new draft AI would do for a set it had no data for. I'm sure the data would come in pretty quickly, but would it just pick randomly until it had the data?

[mcrawford620]

Did any of this make it into the current code base? I don't see it. I was thinking of doing a little work on the BoosterDraftAI but this sounds much more comprehensive, and I won't bother spending much time on BoosterDraftAI if it is going to be replaced.

[Rob Cashwalker]

Nothing significant was completed on this. Anything you want to work on in this area is open.

Determining a card's rating has been left open for debate for a while.

I would like to continue with collecting ratings data locally and aggregated through the server and database. There are some skeletons of this in the code base already. The database schema and PHP can be changed to fit any new ideas.

[mcrawford620]

Got it. My ideas were all just about cleaning up and fine-tuning the logic that is already there. But it sounded like it was going to move to a fully ratings-based approach. Is that what you were thinking?

[mcrawford620]

Well, I started in on making some tweaks. But it definitely seems like just using draft data is going to be the best way in the end. I can't imagine writing a system to evaluate the cards that would produce a better result. I don't have any new ideas other than a straight average draft position.

If you have data already collected and could whip up a PHP page, I could probably copy the system in GuiDownloadPrices and ReadPriceList (?) and use the data in the draft. My only other idea was to use another website's data like http://tappedout.net/mtg-set/2012-core- ... draft-pick but that would be a pain in the butt to crawl and parse.

[Rob Cashwalker]

I might be able to work with that site's data and import into our own database.

[mcrawford620]

Cool. Another option is: http://draft.bestiaire.org/ranking.php if it happens to be easier to work with.
I checked in some tweaks to BoosterDraftAI which will still be useful for when there is no data.

[mcrawford620]

I've been able to pull down a few sets of rankings from bestiaire.org and am just putting it into a txt file for now. I'm working on getting the drafting AI to use it. In the first pass I got the AI to take Doom Blade first which is a nice pick when the Rare in the pack is a land...

[juzamjedi]

As long as that website has good rankings for cube cards it's fine by me. For example, Moxen are insane high picks that AI should take over almost anything.

[mcrawford620]

Can you check out the site yourself and tell me what you think?
http://draft.bestiaire.org/ranking.php
Which cubes would you be thinking of drafting?
My impression is that the site has a lot of data for recent sets, then less data for stuff like Beta or Ice Age, because not many people are online drafting those things. So the Mox stuff in Beta is not ranked 1-5 but still fairly high. Overall, though, as far as I can tell, the data is pretty good.

If you're drafting a custom cube, then it would be comparing ranks across different sets. We might have to adjust for the size of the set at some point so a terrible card from a 150-card small set doesn't get drafted over an average card from a core set.

[delirimouse]

Another option would be for the person who posts a cube to have the option of including a pick order with it as part of the data. It would be a bit of work, but it shouldn't be too hard to come up with. Then the AI could just use the included pick order.

[juzamjedi]

Another option would be for the person who posts a cube to have the option of including a pick order with it as part of the data. It would be a bit of work, but it shouldn't be too hard to come up with. Then the AI could just use the included pick order.

I would guess that many people use the cube list I post / update from time to time. I'm not opposed to providing a pick order as well, but I bet my pick order would be in the same ball park as "highest average pick for people drafting this cube". Some time ago several devs started collecting the data although I have no idea if they have done anything with it yet.

I checked out Cube ranking on the Bestiare website. I see some rankings that I don't think seem right (just as an example, there's no way I'm taking 2 different Garruks over Sol Ring and Moxen). However, if that ranking was somehow easier to use instead of rolling our own I could live with it.

[mcrawford620]

Well, the current system is easier in the sense that it's already in there.

There definitely are sample size issues with some of the rankings. For instance, a recent set like M13 has been drafted on that site thousands of times. Even a Mythic like Garruk has been in nigh on a thousand drafts: http://draft.bestiaire.org/graph_picks. ... d_id=47998
So the rankings for recent sets are fairly stable and established.
While Sol Ring has only been in 9 drafts in Unlimited, and that was an Uncommon back then.
http://draft.bestiaire.org/graph_picks. ... d_id=38607