You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In anticipation of 5.0.0 I wonder if the ResultSet is perhaps worth simplifying a lot. Currently it does a huge amount of calculations (some includes a fair bit of complexity to do all the stuff on disc etc...) that are very specific to the IPD.
I think there are 2 possibilities:
Keep the ResultSet and aim to have it generic so as to do the same amount of calculations for any/all games.
Make the generic output of the Tournament.play() be a payoff matrix and include the option to write the interactions. For the case of the IPD this would lead to no loss of functionality as we would "simply" move the current ResultSet calculations to a specific analysis.ipd spot that we could point at the interactions.
I very much prefer option 2 and wonder if that would streamline things so that we could use the Tournament class in the Moran process? (but this is a minor point)
The text was updated successfully, but these errors were encountered:
Sounds like a good idea to me. For the generic case we could do something similar to #1353 where the output of a tournament is a fully parseable collection of matches and outcomes -- names of players, match specification including the random seed, etc. Then each game can output some additional computations and plots as desired, with some of those perhaps being generic, such as the score distributions (assuming the Game output is real-valued).
In anticipation of
5.0.0
I wonder if theResultSet
is perhaps worth simplifying a lot. Currently it does a huge amount of calculations (some includes a fair bit of complexity to do all the stuff on disc etc...) that are very specific to the IPD.I think there are 2 possibilities:
ResultSet
and aim to have it generic so as to do the same amount of calculations for any/all games.Tournament.play()
be a payoff matrix and include the option to write the interactions. For the case of the IPD this would lead to no loss of functionality as we would "simply" move the currentResultSet
calculations to a specificanalysis.ipd
spot that we could point at the interactions.I very much prefer option 2 and wonder if that would streamline things so that we could use the
Tournament
class in the Moran process? (but this is a minor point)The text was updated successfully, but these errors were encountered: