class CoefficientOptimizer:
"""Class for optimizing Gaussian/reference overlap.
Given matrices of scalar products s and S as returned by overlaps(),
finds the optimal set of coefficients resulting in the largest overlap.
ccount is the number of coefficients.
if fix is True, the first coefficient will be set to 1, and only the
remaining coefficients will be subject to optimization.
"""
def __init__(self, s_kmii, S_kmii, ccount, fix=False):
self.s_kmii = s_kmii
self.S_kmii = S_kmii
self.fix = fix
function = self.evaluate
self.lastterms = None
if fix:
function = self.evaluate_fixed
ccount -= 1
ones = np.ones((ccount, ccount))
diag = np.identity(ccount)
simplex = np.concatenate((np.ones((ccount, 1)), ones + .5 * diag),
axis=1)
simplex = np.transpose(simplex)
self.amoeba = Amoeba(function, simplex, tolerance=1e-10)
def find_coefficients(self):
self.amoeba.optimize()
coefficients = self.amoeba.simplex[0]
if self.fix:
coefficients = [1.] + list(coefficients)
return coefficients
def evaluate_fixed(self, coef):
return self.evaluate([1.] + list(coef))
def evaluate(self, coef):
coef = np.array(coef) # complex coefficients?
terms_km = np.zeros(self.S_kmii.shape[0:2])
for i, (s_mii, S_mii) in enumerate(zip(self.s_kmii, self.S_kmii)):
for j, (s_ii, S_ii) in enumerate(zip(s_mii, S_mii)):
numerator = np.vdot(coef, np.dot(S_ii, coef))
denominator = np.vdot(coef, np.dot(s_ii, coef))
terms_km[i, j] = numerator / denominator
#print terms_km
self.lastterms = terms_km
quality = terms_km.sum()
badness = -quality
return badness
class CoefficientOptimizer:
"""Class for optimizing Gaussian/reference overlap.
Given matrices of scalar products s and S as returned by overlaps(),
finds the optimal set of coefficients resulting in the largest overlap.
ccount is the number of coefficients.
if fix is True, the first coefficient will be set to 1, and only the
remaining coefficients will be subject to optimization.
"""
def __init__(self, s_kmii, S_kmii, ccount, fix=False):
self.s_kmii = s_kmii
self.S_kmii = S_kmii
self.fix = fix
function = self.evaluate
self.lastterms = None
if fix:
function = self.evaluate_fixed
ccount -= 1
ones = np.ones((ccount, ccount))
diag = np.identity(ccount)
simplex = np.concatenate((np.ones((ccount,1)),
ones + .5 * diag), axis=1)
simplex = np.transpose(simplex)
self.amoeba = Amoeba(function, simplex, tolerance=1e-10)
def find_coefficients(self):
self.amoeba.optimize()
coefficients = self.amoeba.simplex[0]
if self.fix:
coefficients = [1.] + list(coefficients)
return coefficients
def evaluate_fixed(self, coef):
return self.evaluate([1.] + list(coef))
def evaluate(self, coef):
coef = np.array(coef) # complex coefficients?
terms_km = np.zeros(self.S_kmii.shape[0:2])
for i, (s_mii, S_mii) in enumerate(zip(self.s_kmii, self.S_kmii)):
for j, (s_ii, S_ii) in enumerate(zip(s_mii, S_mii)):
numerator = np.vdot(coef, np.dot(S_ii, coef))
denominator = np.vdot(coef, np.dot(s_ii, coef))
terms_km[i, j] = numerator / denominator
#print terms_km
self.lastterms = terms_km
quality = terms_km.sum()
badness = - quality
return badness
def __init__(self, s_kmii, S_kmii, ccount, fix=False):
self.s_kmii = s_kmii
self.S_kmii = S_kmii
self.fix = fix
function = self.evaluate
self.lastterms = None
if fix:
function = self.evaluate_fixed
ccount -= 1
ones = np.ones((ccount, ccount))
diag = np.identity(ccount)
simplex = np.concatenate((np.ones((ccount, 1)), ones + .5 * diag),
axis=1)
simplex = np.transpose(simplex)
self.amoeba = Amoeba(function, simplex, tolerance=1e-10)
def __init__(self, s_kmii, S_kmii, ccount, fix=False):
self.s_kmii = s_kmii
self.S_kmii = S_kmii
self.fix = fix
function = self.evaluate
self.lastterms = None
if fix:
function = self.evaluate_fixed
ccount -= 1
ones = np.ones((ccount, ccount))
diag = np.identity(ccount)
simplex = np.concatenate((np.ones((ccount,1)),
ones + .5 * diag), axis=1)
simplex = np.transpose(simplex)
self.amoeba = Amoeba(function, simplex, tolerance=1e-10)
