Closed bmaltais closed 7 years ago
This is basically what combo mode does already.
OK. Reading the code I tought it was just picking one at random:
return model.RandomState(buffer, ShapeType(rnd.Intn(5)+1), a, rnd)
I interpreted this as a single pass based on a random pick of one of the possible mode... not cycling exhaustively through each to find the best fit for the iteration.
I am not seeing the for loop across all possible mode for that option. I might be reading the code wrong.
Very interesting what you have done.
I was wondering if the following mode would not provide even better fit quicker:
This way you would get a blend of shapes that provide the quickest score decrease based on the best shape for the iteration.
Trying to understand the source code so I could try to implement. This could become a new -m option for "optimized blend mode"