Hybris-Python

Installation

To install from this git repo in a virtualenv:

pip install 'hybris-py @ git+https://github.com/Kaeryv/Hybris'

If you are installing this software on a distributed cluster with different architectures, prefer an explicit:

CFLAGS="-march=x86-64 -O2" pip install 'hybris-py @ git+https://github.com/Kaeryv/Hybris'

for maximum compatibility.

Getting started

Optimizing any function

An optimization of function Sphere can be conducted as follows

def objective_function(X):
    return np.mean(np.power(X, 2), axis=-1)

from hybris.optim import ParticleSwarm
opt = ParticleSwarm(20, [10, 0], max_fevals=200)
opt.vmin = -5.0, opt.vmax = 5.0 # Boundaries
opt.reset(456349)

while not opt.stop():
    decision_variables = opt.ask()
    objective_values = objective_function(decision_variables)
    opt.tell(objective_values)

# Show the resulting profile
import matplotlib.pyplot as plt
plt.semilogy(opt.profile)
plt.savefig("Profile.png")

Optimizing the optimizer

To do meta-optimization, any categorical optimizer can be used. We provide a simplified way to do so in the meta module.

from hybris.meta import optimize_self
# Optimizing controls for omega and hybridation with QPSO
prof = optimize_self("1001000", 43)
import matplotlib.pyplot as plt
plt.plot(prof)
plt.show()