Study mode - always using the "average" shuffle

[Hertz Doughnut]

Here's a crazy idea... I was thinking about Dominion simulators... specifically how they play thousands of games and aggregate the results, when I wondered:

What if we had an option where two human players could play an "aggregated" game with each other for the purposes of studying strategies? An aggregated game is the same as a normal game except for one thing... whenever a deck needs to be shuffled, shuffle it, say, 10,000 times, keeping a tally of the deck sequence for each one. Then, continue the game using whichever sequence was the most common.

I think this would be a useful tool for doing the postmortem of a game. Does the money strategy always beat the engine here? Or did it just get lucky?


Keep up the great work!

Kind regards,
HD

[fisherman]

I don't get it. Aren't all the sequences equally likely?

[Hertz Doughnut]

Ah, good point. You're right. The method I wrote about wouldn't do anything.

Let me take another stab at it...

When a deck needs to be shuffled, note the number of cards to be drawn after shuffling. We'll call that n. Shuffle the deck and look at the top n cards. Make a tally of that combination (order doesn't matter). Repeat 10,000 times. Then for the "real" shuffle, we top-deck the combination with the highest tally, and put the rest of the cards in a random order.

So, for example, if we purchased two Ambassadors on T1 and T2, our next hand would be {Copper x 3, Estate, Ambassador} with the rest of the deck random after that. The point being that even though it could happen that {Estate x 3, Ambassador x 2} is your T3 hand, there's little sense in analyzing the strategic implications of something that happens 0.1% of the time.

Make sense?

Thanks for helping me think this through.

Kind regards,
HD

[Stef]

While I can understand what you're trying to accomplish, I don't think its results would be very meaningful.

Very often when trying to build a deck I look for the right balance between "power" and "reliability". Being ahead of my opponent pushes my to make decisions in favor of reliability, while being behind pushes the decisions towards power.

The core of your idea seems to be to remove the "reliability" aspect.

What you'd end up with isn't an analysis of Dominion anymore, it's an entirely different game. Maybe also fun to play, but the results won't say much about Dominion.

[Hertz Doughnut]

Thanks for the reply!  That angle is totally new to me... I'm already finding it helpful to think of the game in terms of power/reliability.

As I've thought more about my request, I think what I'd prefer is a browser extension that turns "study mode" on or off.  With study mode on you'd get a live scoreboard of the following:
- a list of cards in each players' decks
- statistics about the deck (e.g. how many coins does it have? coin density? terminals? villages? buys?)
- the odds of the next 5 cards on top of each deck

Something like that would help me learn how to build engines more efficiently... and judge the situations where engine isn't as good as money.


Thanks, as always, for the great work.  I'm constantly impressed with the quality of this platform.

Kind regards,
HD

[jeebus]

A list of each player's deck at the end of the game is something I hope is in the works anyway.

I think the original idea would definitely be an analysis of Dominion. It would tell you how a strategy works on average, same as the simulators do. The simulators could also be said to remove the "reliability" aspect by the same token. In any case it doesn't sound like a good use of server resources to shuffle thousands of times for a single game analysis.

[TheDetour]

I think the planned Game Logs will suffice for this.  I don't think study mode would be that useful.  Game Logs with all this information in it would be a lot easier to use I would think.