Similar to max-distance, a strategy that is globally optimal several steps into the future would be cool. If the computational burden is too high, we can do a sampling style method ala SKM.
A recursive implementation may be the way to go. To facilitate this, perhaps kaczmarz.Base should accept a normalize parameter which promises the matrix will already have rows with norm 1.
Similar to max-distance, a strategy that is globally optimal several steps into the future would be cool. If the computational burden is too high, we can do a sampling style method ala SKM.
A recursive implementation may be the way to go. To facilitate this, perhaps
kaczmarz.Base
should accept anormalize
parameter which promises the matrix will already have rows with norm 1.