Dominion Strategy Forum

Please login or register.

Login with username, password and session length

Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Messages - InfinitePerplexity

Filter to certain boards:

Pages: [1]
1
Nomad Camp. Woodcutter is rated as the second-worst $3 card, yet somehow Nomad Camp, which is basically the same card is not in the bottom TEN $4 cards? I don't get it. You get the on-gain effect one time, which I guess is useful when you're trying to mass $2 cards asap (i.e. Fool's Gold), and can maybe do cute things with gain-then-draw. But a lot of the time I'd actually rather NOT have the on-buy effect, since when you add your +buy card to an engine, you don't want it in your starting 5 cards.

Yeah, actually, this is puzzling to me, too. Nomad Camp seems so shiny! But it's really just Woodcutter. Not sure why it appears that way to me.

I think this is just an artifact of ranking cards according to cost.  Nomad Camp is at least as bad as Woodcutter, but it so happens that there are more bad $4 cards than there are bad $3 cards, so Nomad Camp ends up further from the bottom.

2
Dominion Articles / Re: The Combo Deck
« on: January 23, 2013, 09:50:47 pm »
Another way to articulate why I don't think we should be calling this deck type the Combo Deck is to point to the Golden Deck as an example.  What's the combo?  Bishop + Gold + Silver?  There are a heck of a lot of things you can substitute for the Gold or the Silver in that combo.  The distinguishing feature of this deck isn't the combination of Bishop with specific other cards, but the use of the Bishop to tightly regulate the composition of your deck and ignore the usual slowdown from buying victory cards.

3
Dominion Articles / Re: The Combo Deck
« on: January 23, 2013, 09:35:49 pm »
I would like to suggest another way of looking at this deck type, instead of focusing on combos and substitutability.  It seems to me that the key feature of many of these deck types is...hmmm..."deck control"?  I'm struggling to find the right way to articulate this.  Apothecary/NV, Golden Deck, Golem/Counting House, and the various pins all seek ways to circumvent the usual mechanic whereby the buying power of your deck is improved or degraded by gaining new actions, treasures, or victory cards.

NV/Bridge and Tactician/Vault seem a bit different to me - more like degenerate cases of engine decks.  In NV/Bridge, for example, both cards are playing something like their usual mega-turn-engine enabling roles, but taken to a ridiculous extreme.

4
Game Reports / Worst Black Market game evar!
« on: March 04, 2012, 02:22:57 am »
Has this topic been done before?  Anyway, this one:

http://dominion.isotropic.org/gamelog/201203/03/game-20120303-231304-da7247aa.html

Pulling something your opponent can't get out of Black Market: Priceless.  Playing it thrice: Tri-Priceless!

5
Game Reports / Re: Dear My Opponent: I am Sorry
« on: March 04, 2012, 02:17:32 am »
No Swindlers in that game?  The curses all came from Followers.

6
Game Reports / Re: Dear My Opponent: I am Sorry
« on: March 03, 2012, 09:44:26 pm »
Dear dr. girlfriend,

I'm sorry you opened 2/5 on this board:

http://dominion.isotropic.org/gamelog/201203/03/game-20120303-183627-5134ced7.html

I got to open Tournament/Steward while you were stuck with Tournament/nothing.  Good on you for avoiding Counting House/nothing.

7
Simulation / Re: machine learning on Isotropic logs
« on: January 25, 2012, 01:12:20 am »
The model does two kinds of output, once you've trained it.  One displays something about the learned buy rules for a specific card; for the ad hoc decision tree classifier I wrote, it gives something like this:

>>> print mymodel["Ironworks"]
* Gardens:
  * Bishop:
    - No Buy (16%)
  * No Bishop:
    * Monument:
      - No Buy (16%)
    * No Monument:
      * Young Witch:
        - No Buy (28%)
      * No Young Witch:
(snipped for readability, goes on to basically say you usually buy Ironworks if Gardens is around)
* No Gardens:
  * Great Hall:
    * Trade Route:
        - Buy (60%)
    * No Trade Route:
(snipped for readability)

The other thing it does is it takes a list of kingdom cards, converts it to a boolean input vector, runs the appropriate ten classifiers to produce an output vector, which it then converts back to something human readable:

>>> mymodel.classify("Highway, Nomad Camp, Harvest, Cartographer, Golem, Talisman, Vineyard, Island, Haggler, and University.")

'Cartographer', 'Island', 'Haggler', 'Talisman', 'Nomad Camp', 'Highway'
{'Golem': 0.083333333333333329, 'Cartographer': 0.90000000000000002, 'Island': 0.72727272727272729, 'University': 0.14285714285714285, 'Haggler': 0.66666666666666663, 'Vineyard': 0.4375, 'Talisman': 0.5625, 'Nomad Camp': 0.97222222222222221, 'Harvest': 0.023809523809523808, 'Highway': 0.69999999999999996}

...which is a mix of good and bad advice, to say the least.  My gut says the approach of predicting each card separately is fundamentally flawed, and I have some ideas for how to improve on it, but I haven't had the chance to try them out.

@rspeer, what sort of input and output did you guys try?

8
Simulation / machine learning on Isotropic logs
« on: January 23, 2012, 12:25:13 am »
Hi all, I've been teaching myself how to do machine learning in Python lately, using the NumPy package and log data downloaded from Isotropic.  My goal is to write a "kingdom analyzer" that looks at a ten-card kingdom and gives advice that would help a beginning or intermediate player understand what sort of deck the collective wisdom of skilled Isotropic players would build for that kingdom.

I've come up with a sort of "minimum viable implementation" and tested it on five days of game logs.  Currently it does a good job picking two-and-three card combos and counters (e.g. buy Ironworks if Garden is available, but not if Bishop is also available), and it could probably find larger combos if I downloaded more data.  But it's poor at designing a coherent deck, and it still makes "wtf" recommendations pretty regularly (I suspect the model I'm using overfits severely.)  Of course, I'm just a level 18, so maybe the model is right and I'm wrong :)

Anyway, I'm curious to see if anyone else has done experiments along these lines, or would like to share thoughts and strategies.  My current model puts the classifiers themselves into wrappers, so it would be pretty easy to swap out algorithms.  I may put the code up on GitHub, although (1) I code like a scientist, so parts of my code are pretty bad and (2) I've never put anything up on GitHub before so I'm not really familiar with how it works. 

Pages: [1]

Page created in 0.09 seconds with 19 queries.