def _get_jacobian(self):
jac = []
for param in self.params:
# Differentiate to every param
f = sympy.diff(self.model, param)
# Make them into pythonic functions
jac.append(sympy_to_scipy(f, self.vars, self.params))
return jac
def __init__(self, model, *args, **kwargs):
"""
:param model: sympy expression.
:param x: xdata to fit to. NxM
:param y: ydata Nx1
"""
super(BaseFit, self).__init__(*args, **kwargs)
self.model = model
# Get all parameters and variables from the model.
self.vars, self.params = seperate_symbols(self.model)
# Compile a scipy function
self.scipy_func = sympy_to_scipy(self.model, self.vars, self.params)
def jacobian(self):
"""
Get the scipy functions for the Jacobian. This returns functions only, not values.
:return: array of derivative functions in all parameters, not values.
"""
if not self.__jac:
self.__jac = []
for param in self.params:
# Differentiate to every param
f = sympy.diff(self.model, param)
# Make them into pythonic functions
self.__jac.append(sympy_to_scipy(f, self.vars, self.params))
return self.__jac
def make_jac(constraint_lhs, p, x):
"""
:param constraint_lhs: equation of a constraint. The lhs is assumed to be an eq, rhs a number.
:param p: current value of the parameters to be evaluated.
:return: numerical jacobian.
"""
sym_jac = []
for param in self.params:
sym_jac.append(sympy.diff(constraint_lhs, param))
ans = np.array(
[sympy_to_scipy(jac, self.vars, self.params)(x, p) for jac in
sym_jac]
)
return ans
def get_constraints(self):
"""
Turns self.constraints into a scipy compatible format.
:return: dict of scipy compatile statements.
"""
from sympy import Eq, Gt, Ge, Ne, Lt, Le
cons = []
types = {
Eq: 'eq', Gt: 'ineq', Ge: 'ineq', Ne: 'ineq', Lt: 'ineq', Le: 'ineq'
}
def make_jac(constraint_lhs, p, x):
"""
:param constraint_lhs: equation of a constraint. The lhs is assumed to be an eq, rhs a number.
:param p: current value of the parameters to be evaluated.
:return: numerical jacobian.
"""
sym_jac = []
for param in self.params:
sym_jac.append(sympy.diff(constraint_lhs, param))
ans = np.array(
[sympy_to_scipy(jac, self.vars, self.params)(x, p) for jac in
sym_jac]
)
return ans
for key, constraint in enumerate(self.constraints):
# jac = make_jac(c, p)
cons.append({
'type': types[constraint.__class__],
# Assume the lhs is the equation.
'fun': lambda p, x, c: sympy_to_scipy(c.lhs, self.vars, self.params)(x, p),
# Assume the lhs is the equation.
'jac' : lambda p, x, c: make_jac(c.lhs, p, x),
'args': (self.xdata, constraint)
})
cons = tuple(cons)
return cons
