Particle Swarm Always Converges After One Refinement; Subsequent Refinements Are Not Different

[Andrew W L Dickinson]

I've observed that PSO always converges after one refinement pass; never have I had a refinement further refined by a second pass. I've always attributed this to the data I've been providing being near-optimal, but after receiving feedback from another with the same observation, I wanted to create an issue to generate discussion.

[Former user]

Yup, that's common (and hopefully will be fixed soon). If the result does not change after 15 epochs, I stop the optimization. I put that because it usually takes ~3 seconds per epoch on my machine. I have some ideas to fix this issue though.