In class he mentioned a problem with loss of diversity and sample variance when resampling. I figured that we might as well just implement low-variance sampling since it's about the same amount of work.
Regular sampling involves generating N random numbers in [0, 1]. In low-variance sampling, we instead generate a random number r in [0, 1 / N) and then use {r + 1 / N ... r + (N - 1) / N} as the rest of the samples.
slide (TODO: there's a better side in a different deck that hasn't been made public yet)
The class outlined in #37 allows us to map an arbitrary CDF value in [0, 1] to a particle.
In class he mentioned a problem with loss of diversity and sample variance when resampling. I figured that we might as well just implement low-variance sampling since it's about the same amount of work.
Regular sampling involves generating N random numbers in [0, 1]. In low-variance sampling, we instead generate a random number
rin [0, 1 / N) and then use {r + 1 / N ... r + (N - 1) / N} as the rest of the samples.slide (TODO: there's a better side in a different deck that hasn't been made public yet)
The class outlined in #37 allows us to map an arbitrary CDF value in [0, 1] to a particle.