Hey all,
working on my implementation of the PSO for an Auto-Tuning-Framework, I was wondering what the implementation of the PSO in OpenTuner is based on?
If I understand your implementation correctly, you're using weights c, c1 and c2 which are all equal to 0.5 scaled by a random float between 0.0 and 1.0. However what I found in theoretical references (e.g. Eberhart and Shi "Tracking and optimizing dynamic systems with particle swarms.") as a good choice for the weights was U(0,1) or even the use of a constricting coefficient instead of weights.
Is there a specific reason you decided to implement the PSO the way it currently is?
Hey all, working on my implementation of the PSO for an Auto-Tuning-Framework, I was wondering what the implementation of the PSO in OpenTuner is based on?
If I understand your implementation correctly, you're using weights
c
,c1
andc2
which are all equal to0.5
scaled by a random float between0.0
and1.0
. However what I found in theoretical references (e.g. Eberhart and Shi "Tracking and optimizing dynamic systems with particle swarms.") as a good choice for the weights wasU(0,1)
or even the use of a constricting coefficient instead of weights.Is there a specific reason you decided to implement the PSO the way it currently is?
Cheers