pycgdescent_example.py

#pele and PyCG_DESCENT need to be installed and added to the pythonpath
import numpy as np
from pele.potentials import BasePotential
from PyCG_DESCENT import CGDescent

class squareFunc(BasePotential):
    # this is the only method that you actually need to implement                                                                                              
    def getEnergy(self, x):
        val = x[0]**2 + x[1]**2
        return val

    # this is optional, if you know the analytical derivative implement it                                                                                     
    # otherwise the base class will automatically compute the gradient numerically
    # using finite differences
    def getEnergyGradient(self, x):
        val = x[0]**2 + x[1]**2
        grad = np.array([2*x[0], 2*x[1]])
        return val, grad

def main():
    f = squareFunc() #initialize the cost function (or potential in our language)
    
    x_start = np.array([1, 1])
    e_start = f.getEnergy(x_start)

    cgd = CGDescent(x_start, f, tol=1e-5, M=0, print_level=0, verbosity=0)
    result = cgd.run(1000) #1000 is the maximum number of steps, it terminates when gradient inf-norm is less than tol
    x_end, e_end = result.coords, result.energy

    print "f(x_start)={} --> f(x_end)={}".format(e_start, e_end)

if __name__=='__main__':
    main()