chanshing / xnes

Implementation of "Exponential Natural Evolution Strategies" (xNES) https://arxiv.org/abs/1106.4487

Implementation of "Exponential Natural Evolution Strategies" (xNES)

https://arxiv.org/abs/1106.4487

Usage

xnes = XNES(f, mu, amat)
xnes.step(1000)
print xnes.mu_best 

where

f fitness function (real-valued function)

mu initial guess of center (scalar or vector)

amat initial guess of covariance matrix (scalar or matrix)

See xnes.py for a specific example.

Notes:

• Adaptation sampling (use_adasam=True) requires tuning the etas a bit to work well. First try without it.
• When using n_jobs>1, it is better to turn off multithreading: export MKL_NUM_THREADS=1
• n_jobs>1 is normally better only if f is super expensive.