Closed mikewolfe closed 2 years ago
Simulated annealing seems to be working best, but we can do replica exchange and particle swarm if desired. They take longer and don't yield adjusted mutual information values as high as simple simulated annealing on test data, though.
Here we plan to create a custom swarm (https://machinelearningmastery.com/a-gentle-introduction-to-particle-swarm-optimization/) or multi-swarm optimization (https://ieeexplore.ieee.org/document/6256566) algorithm to determine the best motifs.
This will go in its own crate https://github.com/mikewolfe/DNAshape_motif_finder/tree/rust_dev/rust_utils/optim