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
Hey, I was wondering whether you actually used the data written to the innovation record at some point. In the current implementation it looks like you are just using it as a counter (and in a doc-string it is noted, that it might be used for checking whether an edge with the same node pair (src, dest) already exists. For getting a new node/edge id count from itertools could be used. For checking whether the node pair exists, a native python set might be more efficient.
Since the innovation record also stores whether a new node was generated and in which generation the innovation first occurred - it seems like you were planning to use that information somehow, but I can't reference in the code itself.
Did you use the information during debugging and analysis of executed experiments? If so, how did you utilize it?
The text was updated successfully, but these errors were encountered:
Hey, for WANNs I did not. The innovation numbers are only used for crossover (which I didn't used) and speciation (which I didn't use).
The ordering can be used in NEAT to set different coefficients for 'compatibility distance', so that when comparing two networks you can set different distances between networks if the different genes are older or younger (comparing to a direct ancestor might be closer than comparing to a cousin). In practice I don't know if I've ever seen anybody set them different -- it is an interesting idea, but with all of the hyperparameters already in NEAT we really don't need any more.
Hey, I was wondering whether you actually used the data written to the innovation record at some point. In the current implementation it looks like you are just using it as a counter (and in a doc-string it is noted, that it might be used for checking whether an edge with the same node pair (src, dest) already exists. For getting a new node/edge id
count
fromitertools
could be used. For checking whether the node pair exists, a native python set might be more efficient.Since the innovation record also stores whether a new node was generated and in which generation the innovation first occurred - it seems like you were planning to use that information somehow, but I can't reference in the code itself.
Did you use the information during debugging and analysis of executed experiments? If so, how did you utilize it?
The text was updated successfully, but these errors were encountered: