Closed maxkw closed 8 years ago
Should have read this before doing my work today. This is captured in my new abstraction for games/decisions. DecisionObserved is a normal decision that calls observe_k on the given observers whenever a decision is made. You can string these together as DecisionSeqs such that a sequence of decisions are played out, where after each decision the observations are processed. To make a game you can string normal Decisions together as DecisionSeqObserveds which play the sequence of decisions and process the observations at the end of everything.
Is this what we wanted?
Yes I think this is good... There are really two notions of observability. The first is what I believe you implemented in your games code. Where we hide observations until a later time in order to study symmetric games etc.
The second (which is the goal of this ticket) is really a property of the world NOT a property of the game. Who in the world observes the game itself i.e., is it a private interaction or does the whole community see what the players did. You might think about a network representation here...
Multi-Agent experiment, Private info vs Public info.
Cool! Can you also make graphs for notnice_RA? Very intuitive results!
On Tue, Jul 19, 2016 at 8:46 PM, Alejandro Vientós <notifications@github.com
wrote:
Multi-Agent experiment, Private info vs Public info. [image: screenshot from 2016-07-19 20-39-22] https://cloud.githubusercontent.com/assets/1637297/16971415/8c477706-4df1-11e6-87b2-2e7842835885.png
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/maxkw/evolution/issues/12#issuecomment-233808435, or mute the thread https://github.com/notifications/unsubscribe-auth/ABQt1Z-4FTb5zStJBKy8vEIWXsFdP65yks5qXW_ogaJpZM4JChcY .
RA: Private v. Public
implement a new run function (this is before generalizing to all the patterns). This should run by generating a single interaction and then having the observation being observed by EVERYONE. Only after you've implemented this parallel run function is it worth it to try generalizing to a one size fits all run function.