def rosen_obj(params, shift):
val = rosen(params["x_half_a"] -
shift) + rosen(params["x_half_b"] - shift)
dval = OrderedDict([
("x_half_a", rosen_der(params["x_half_a"] - shift)),
("x_half_b", rosen_der(params["x_half_b"] - shift)),
])
return val, dval
def compute_dr_wrt(self, wrt):
if wrt is self.x:
if visualize:
import matplotlib.pyplot as plt
residuals = np.sum(self.r**2)
print('------> RESIDUALS %.2e' % (residuals,))
print('------> CURRENT GUESS %s' % (str(self.x.r),))
plt.figure(123)
if not hasattr(self, 'vs'):
self.vs = []
self.xs = []
self.ys = []
self.vs.append(residuals)
self.xs.append(self.x.r[0])
self.ys.append(self.x.r[1])
plt.clf();
plt.subplot(1,2,1)
plt.plot(self.vs)
plt.subplot(1,2,2)
plt.plot(self.xs, self.ys)
plt.draw()
return row(rosen_der(self.x.r))
def compute_dr_wrt(self, wrt):
if wrt is self.x:
if visualize:
import matplotlib.pyplot as plt
residuals = np.sum(self.r**2)
print '------> RESIDUALS %.2e' % (residuals,)
print '------> CURRENT GUESS %s' % (str(self.x.r),)
plt.figure(123)
if not hasattr(self, 'vs'):
self.vs = []
self.xs = []
self.ys = []
self.vs.append(residuals)
self.xs.append(self.x.r[0])
self.ys.append(self.x.r[1])
plt.clf();
plt.subplot(1,2,1)
plt.plot(self.vs)
plt.subplot(1,2,2)
plt.plot(self.xs, self.ys)
plt.draw()
return row(rosen_der(self.x.r))
def testEstimation(self):
values = np.random.uniform(-5.0, 5.0, size=[10, 2])
for i in range(10):
true_grad = rosen_der(values[i, :])
pred_grad = self.estimator.simpleFiniteDifferences(values[i, :])
self.assertAlmostEqual(true_grad[0], pred_grad[0], 4)
self.assertAlmostEqual(true_grad[1], pred_grad[1], 4)
def testMaximization(self):
self.optimizer.isMaximize = True
rosen_inv = lambda x: - rosen(x)
rosen_der_inv = lambda x: - rosen_der(x)
max_params, max_value, _ = self.optimizer.optimize(rosen_inv, rosen_der_inv, x0=self.x0)
self.assertAlmostEqual(self.reference_value, max_value, 10)
self.assertAlmostEqual(self.reference_params[0], max_params[0], 4)
self.assertAlmostEqual(self.reference_params[1], max_params[1], 4)
def nltest(x,grad):
nonlocal counter
nonlocal countergrad
if len(grad) > 0:
countergrad += 1
grad[:] = rosen_der(x)
return grad
counter += 1
return rosen(x)
def test_optimization_trust():
print("**********************************************")
print("TEST Newton trust region ")
x0 = [1.3, 0.7, 0.8, 1.9, 1.2]
res = optimize.minimize(
optimize.rosen, x0, method='trust-ncg',
jac=optimize.rosen_der,
hess=optimize.rosen_hess,
options={'gtol': 1e-8, 'disp': True})
print(res.x)
print(optimize.rosen(x0).shape)
print(optimize.rosen_der(x0).shape)
print(optimize.rosen_hess(x0).shape)
return res.fun
def test_gradient():
""" Test the gradient using rosen and higher order functions
"""
def F(x):
return x[0]**2 + x[0]*x[1] + x[1]**2
def F_grad(x):
return array([2*x[0]+x[1],x[0]+2*x[1]])
opt1 = p.OptimizationProblem(rosen)
opt2 = p.OptimizationProblem(F)
for i in range(-3,3):
for j in range(-3,3):
k = opt1.gradient([float(i),float(j)]) - \
rosen_der([float(i),float(j)])
kk = opt2.gradient([float(i),float(j)]) - \
F_grad([float(i),float(j)])
assert(sum( abs((k+kk)) <1e-5 )==2)
def rosen_obj_func_grad(w):
return sparse.csc_matrix(rosen_der(w.T.toarray()[0])).T
def rosen_obj(params):
val = rosen(params["x"])
dval = OrderedDict()
dval["x"] = rosen_der(params["x"])
return val, dval
def rosen_der_wrapper(self, x, args=()):
self.ngev += 1
return rosen_der(x, *args)
def fprime(self, x):
return rosen_der(x)
eps = 1.0e-5
xtol = 1.0e-16
icall = 0
iflag = 0
# initial guess
x = np.zeros(n)
for i in range(0, n, 2):
x[i] = -1.2
x[i + 1] = 1.0
x0 = x.copy()
# initial evaluation
fval = 0.0
gval = np.zeros_like(x)
fval = rosen(x)
gval = rosen_der(x)
print("initial function value = {}".format(fval))
print("initial gradient norm = {}".format(np.sqrt(np.dot(gval, gval))))
for icall in range(2000):
[xk, oflag] = lbfgs.lbfgs(n=n, m=m, x=x, f=fval, g=gval, \
diagco=diagco, diag=diag, \
iprint=iprint, eps=eps, xtol=xtol, w=work, iflag=iflag)
iflag = oflag
x = xk[:]
fval = rosen(x)
gval = rosen_der(x)
print("iflag = {}".format(iflag))
#print("x - x0 = {}".format(x-x0))
#print("diag = {}".format(diag))
#print("current function value = {}".format(fval))
def gradient(x, data=None):
return rosen_der(x)
def grad(self, x):
g = rosen_der(x)
return g
def gradient(x, data):
return rosen_der(x)
def gradient(self):
inp = self._position.to_global_data_rw()
out = ift.Field.from_global_data(space, rosen_der(inp))
return out
def rosen_der_wrapper(self, x, args=()):
self.ngev += 1
return rosen_der(x, *args)
def _getGradient(self, x, *args):
return rosen_der(x)
def grad_rosen(xx):
yy = xx.flatten()
gra = spo.rosen_der(yy)
return gra.reshape((gra.shape[0], 1))
def f(x, g):
g[:] = rosen_der(x)
print "one call"
return rosen(x)
def fp(x):
return rosen_der(x).reshape(3, 1)
def rosenbrock_function_jacobian(samples):
assert samples.shape[1] == 1
return rosen_der(samples).T
def rosen_func_and_der(x):
return rosen(x), rosen_der(x)
def SensEq(x, f, g, gc):
dfdx = rosen_der(x)
return dfdx, []
def rosenbrock_grad_f(x):
return rosen_der(x)
def rosen_der_inv(x):
return - rosen_der(x)
def fprime(self, x):
return rosen_der(x)
def callback(xk):
print(xk)
print(rosen_der(xk["x"]))
dr = rosen_der(xk["x"])
log.append([xk["x"][0], xk["x"][1], -lr * dr[0], -lr * dr[1]])
def SensEq2(p, r, g, x):
return rosen_der(p)
def func_grad_hess(x, *args):
f = optimize.rosen(x)
g = optimize.rosen_der(x)
h = optimize.rosen_hess(x)
return (f, g, h)
def cost_fun(params, **kwargs):
return rosen_der([params['a'], params['b']])
def func_grad_hess(x,*args):
f = optimize.rosen(x)
g = optimize.rosen_der(x)
h= optimize.rosen_hess(x)
return (f,g,h)
def f(x):
time.sleep(0.01)
return [opt.rosen(x), opt.rosen_der(x)]
def f(x):
time.sleep(0.1)
return [opt.rosen(x), opt.rosen_der(x)]