def _setupDynamicsConstraints(self,endTime,traj):
# Todo: add parallelization
# Todo: get endTime right
g = []
nicp = 1
deg = 4
p = self._dvMap.pVec()
for k in range(self.nk):
newton = Newton(LagrangePoly,self.dae,1,nicp,deg,'RADAU')
newton.setupStuff(endTime)
X0_i = self._dvMap.xVec(k)
U_i = self._U[k,:].T
# guess
if traj is None:
newton.isolver.setOutput(1,0)
else:
X = C.DMatrix([[traj.lookup(name,timestep=k,degIdx=j) for j in range(1,traj.dvMap._deg+1)] \
for name in traj.dvMap._xNames])
Z = C.DMatrix([[traj.lookup(name,timestep=k,degIdx=j) for j in range(1,traj.dvMap._deg+1)] \
for name in traj.dvMap._zNames])
newton.isolver.setOutput(C.veccat([X,Z]),0)
_, Xf_i = newton.isolver.call([X0_i,U_i,p])
X0_i_plus = self._dvMap.xVec(k+1)
g.append(Xf_i-X0_i_plus)
return g
def _setupDynamicsConstraints(self, endTime, traj):
# Todo: add parallelization
# Todo: get endTime right
g = []
nicp = 1
deg = 4
p = self._dvMap.pVec()
for k in range(self.nk):
newton = Newton(LagrangePoly, self.dae, 1, nicp, deg, 'RADAU')
newton.setupStuff(endTime)
X0_i = self._dvMap.xVec(k)
U_i = self._U[k, :].T
# guess
if traj is None:
newton.isolver.setOutput(1, 0)
else:
X = C.DMatrix([[traj.lookup(name,timestep=k,degIdx=j) for j in range(1,traj.dvMap._deg+1)] \
for name in traj.dvMap._xNames])
Z = C.DMatrix([[traj.lookup(name,timestep=k,degIdx=j) for j in range(1,traj.dvMap._deg+1)] \
for name in traj.dvMap._zNames])
newton.isolver.setOutput(C.veccat([X, Z]), 0)
_, Xf_i = newton.isolver.call([X0_i, U_i, p])
X0_i_plus = self._dvMap.xVec(k + 1)
g.append(Xf_i - X0_i_plus)
return g
def step(self,t,u,h): system = self.system noise = np.random.normal(size=[len(system.noise(t,u).T)]) def residual(v): return (system.mass(t+h,v) - system.mass(t,u))/h - system.deterministic(t+h,v) - np.sqrt(h)/h*np.dot(system.noise(t,u),noise) N = Newton(residual) ## N.xtol = min(N.xtol, h*1e-4) result = N.run(u) return t+h, result
def step(self, t, u, h):
system = self.system
noise = np.random.normal(size=[len(system.noise(t, u).T)])
def residual(v):
return (
(system.mass(t + h, v) - system.mass(t, u)) / h
- system.deterministic(t + h, v)
- np.sqrt(h) / h * np.dot(system.noise(t, u), noise)
)
N = Newton(residual)
## N.xtol = min(N.xtol, h*1e-4)
result = N.run(u)
return t + h, result
def _setupDynamicsConstraints(self):
# Todo: add parallelization
# Todo: add initialization
g = []
nicp = 10
deg = 4
p = self._dvMap.pVec()
for k in range(self.nk):
newton = Newton(LagrangePoly,self.dae,1,nicp,deg,'RADAU')
endTime = 0.05
newton.setupStuff(endTime)
X0_i = self._dvMap.xVec(k)
U_i = self._dvMap.uVec(k)
_, Xf_i = newton.isolver.call([X0_i,U_i,p])
X0_i_plus = self._dvMap.xVec(k+1)
g.append(Xf_i-X0_i_plus)
return g
def _setupDynamicsConstraints(self):
# Todo: add parallelization
# Todo: add initialization
g = []
nicp = 10
deg = 4
p = self._dvMap.pVec()
for k in range(self.nk):
newton = Newton(LagrangePoly, self.dae, 1, nicp, deg, 'RADAU')
endTime = 0.05
newton.setupStuff(endTime)
X0_i = self._dvMap.xVec(k)
U_i = self._dvMap.uVec(k)
_, Xf_i = newton.isolver.call([X0_i, U_i, p])
X0_i_plus = self._dvMap.xVec(k + 1)
g.append(Xf_i - X0_i_plus)
return g
class TestGaussianClass(unittest.TestCase):
def setUp(self):
self.newton = Newton(f='x**2 - 9', max_iter=1e6)
def test_initialization(self):
self.assertEqual(self.newton.f, 'x**2 - 9', 'incorrect function')
self.assertEqual(self.newton.max_iter, 1e6,
'incorrect maximum number of iterations')
def test_solution1calc(self):
self.assertEqual(round(self.newton.find_solution(1000), 2), 3.00,
'root incorrect')
def test_solution2calc(self):
self.assertEqual(round(self.newton.find_solution(-1000), 2), -3.00,
'root incorrect')
def test_handlingZeroDivision(self):
self.assertEqual(round(self.newton.find_solution(0), 2), 3.00,
'Zero Division error occured')
def test_simple_surface():
# the class of test surface
PES = SimpleSurface()
# dimer algorithm
ini_position = (np.random.rand(2) - 0.5) * 10
ini_vector = np.random.rand(2)
d = Dimer(2, ini_position, ini_vector, whether_print=False)
position_d, times_d = d.work() # 得到dimer运行轨迹和每一次的旋转数
# print('vector', d.vector)
# quasi newton method
qN = Newton(PES.get_value, PES.get_diff,
[position_d[-1, 0], position_d[-1, 1]])
position_qN, times_qN = qN.bfgs_newton(PES.get_hess)
position = np.concatenate((position_d, position_qN), axis=0)
PES.show_point_2d(position)
plt.plot(position_d[-1, 0], position_d[-1, 1], 'ko')
PES.show_surface_2d(-5, 5)
plt.title('Dimer rotates %d times and run %d times \n '
'Newton runs %d times' % (sum(times_d), len(times_d), times_qN))
plt.show()
def solve(x0, method, oracle, visual=False):
time_start = process_time()
if method == "BFGS":
F, x, G = bfgs(oracle, x0, visual=visual)
if method == "Newton":
F, x, G = Newton(oracle, x0, visual=visual)
if method == "PR":
F, x, G = Polak_Ribiere(oracle, x0, visual=visual)
if method == "Gradient_F":
F, x, G = Gradient_F(oracle, x0, visual=visual)
cpu_time = process_time() - time_start
print("Temps d'exécution : ", cpu_time)
return F, x, G
def run(step, point_type, lambda_p, method):
f = Function(lambda_p)
n = f.get_size() # x size
# Initial point
if point_type == "const":
x0 = np.ones(n)
else:
x0 = np.random.uniform(-2, 2, n)
mxitr = 50000 # Max number of iterations
tol_g = 1e-8 # Tolerance for gradient
tol_x = 1e-8 # Tolerance for x
tol_f = 1e-8 # Tolerance for function
# Method for step update
if step == "fijo":
msg = "StepFijo"
elif step == "hess":
msg = "StepHess"
else:
msg = "Backtracking"
step_size = 1 # Gradient step size for "StepFijo" method
# Estimate minimum point through optimization method chosen
if method == "gd":
alg = GD()
elif method == "newton":
alg = Newton()
else:
print("\n Error: Invalid optimization method: %s\n" % method)
return
xs = alg.iterate(x0, mxitr, tol_g, tol_x, tol_f, f, msg, step_size)
# Print point x found and function value f(x)
# print("\nPoint x found: ", xs[-1])
print("\nf(x) = ", f.eval(xs[-1]))
plt.plot(np.array(range(n)), xs[-1])
plt.plot(np.array(range(n)), f.y)
plt.legend(['x*', 'y'], loc = 'best')
plt.show()
def newton_results():
iterastr = request.form.get("itera")
x0str = request.form.get("x0")
funstr = request.form.get("fun")
dfunstr = request.form.get("dfun")
tolstr = request.form.get("tol")
if iterastr == '' or x0str == '' or funstr == '' or dfunstr == '' or tolstr == '':
return "<h1 style='text-align: center;'>You are missing one or more values</h1>"
try:
itera = int(iterastr)
except:
return "<h1 style='text-align: center;'>Check iteration value</h1>"
try:
x0 = float(x0str)
except:
return "<h1 style='text-align: center;'>Check initial value</h1>"
try:
fun = sp.sympify(funstr)
except:
return "<h1 style='text-align: center;'>Check function entered</h1>"
try:
dfun = sp.sympify(dfunstr)
except:
return "<h1 style='text-align: center;'>Check </h1>"
try:
tol = sp.sympify(tolstr)
except:
return "<h1 style='text-align: center;'>Check tolerance entered {{tol}}</h1>"
try:
list_a, list_f, list_e, root = Newton(itera, x0, fun, dfun, tol)
except:
return "<h1 style='text-align: center;'>Check values entered</h1>"
funplot = funstr.replace("**", "^")
it = len(list_a)
list_it = list(range(0, it))
return render_template("newton.html",
list_a=list_a,
list_f=list_f,
list_e=list_e,
list_it=list_it,
root=root,
funplot=funplot)
def newton_results():
itera = request.form.get("itera")
x0 = request.form.get("x0")
fun = request.form.get("fun")
dfun = request.form.get("dfun")
tol = request.form.get("tol")
if itera == '' or x0 == '' or fun == '' or dfun == '' or tol == '':
return "<h1 style='text-align: center;'>Check Values Entered</h1>"
list_a, list_f, list_e, root = Newton(itera, x0, fun, dfun, tol)
it = len(list_a)
list_it = list(range(0, it))
return render_template("newton.html",
list_a=list_a,
list_f=list_f,
list_e=list_e,
list_it=list_it,
root=root)
def test_Newton():
from sympy import symbol, diff, lambdify
x = symbol.Symbol('x') # определяем математический символ x
f_expr = x**2 - 4 # символьное выражение для f(x)
dfdx_expr = diff(f_expr, x) # вычисляем символьно f'(x)
# Преобразуем f_expr и dfdx_expr в обычные функции Python
f = lambdify(
[x], # аргумент f
f_expr)
dfdx = lambdify([x], dfdx_expr)
tol = 1e-3
computed, no_iterations = Newton(f, dfdx, x=1, eps=tol)
expected = 2.
success = np.abs(computed - expected) <= tol
msg = 'expected = {},\n computed = {}, Число итераций = {}'.format(
expected, computed, no_iterations)
assert success, msg
import sys
import itertools
# Mainlib
from score_regressor import get_best
from newton import Newton
from newton.knn import *
from newton.forest import *
app = Flask(__name__)
CORS(app)
setattr(sys.modules['__main__'], 'gower_distance', gower_distance)
MODEL_CACHE = 5
sisrec = Newton(np.array([i for i in range(1, 12)]),
'newton/forest/serialized', 'newton/knn/serialized',
'newton/data', MODEL_CACHE, 4, 5)
@app.route("/get_prediction", methods=["GET", "POST", "PUT"])
def get_predicition():
# Para evitar problemas con Chrome
# resp.headers['Access-Control-Allow-Origin'] = '*'
# Data viene en formato json -> {}
# {uid}
if request.method == "GET":
result = 'GET request is not allowed'
elif request.method == "POST":
# Diccionario de data que viene en el request
from newton import Newton f = '2*x**2 - 50' newton = Newton(f) newton.find_solution(1)
def get_random_uniform_in_range(x_rng: np.array, y_rng: np.array) -> np.array:
p = np.random.rand(2)
p[0] *= x_rng[1] - x_rng[0]
p[0] += x_rng[0]
p[1] *= y_rng[1] - y_rng[0]
p[1] += y_rng[0]
return p
if __name__ == "__main__":
opt = Newton(obj_func=obj_func,
gradient_func=gradient,
reg_inv_hessian=reg_inv_hessian)
opt.log.setLevel(logging.INFO)
x_rng = [-2, 2]
y_rng = [-1, 1]
fig_dir = "figures"
Path(fig_dir).mkdir(parents=True, exist_ok=True)
# params_init = get_random_uniform_in_range(x_rng, y_rng)
# params_init = np.array([-1.55994695, -0.31833122])
params_init = np.array([-1.1, -0.5])
converged, no_opt_steps, final_update, traj, line_search_factors = opt.run(
no_steps=10, params_init=params_init, tol=1e-8, store_traj=True)
def f(x):
return x**2 + 4 * np.sin(x)
left = int(input('Enter left border: '))
right = int(input('Enter right border: '))
n = int(input('Enter number of points: '))
bf = Bruteforce(left, right, n, f)
bf.FirstStep()
bf.plots()
s1 = Secant.find(f, a=-2, b=-1)
s2 = Secant.find(f, a=-1, b=1)
b1 = Bisect.find(f, a=-2, b=-1)
b2 = Bisect.find(f, a=-1, b=1)
print("\nSecant: {} {}".format(s1, s2))
print("Bisect: {} {}".format(b1, b2))
def dfunc(x):
return 2 * x + 4 * cos(x)
newton = Newton(f, dfunc, x=2, max_iterations=50, eps=10e-7)
root = newton.calculate()
print("Newton: {}".format(root))
def setUp(self):
self.newton = Newton(f='x**2 - 9', max_iter=1e6)
from newton import Newton
# Ejercicio 1
newton = Newton(
value_range=[0, 10],
f=lambda l: l * (20 - 2 * l)**2,
fp=lambda l: 12 * l**2 - 160 * l + 400,
fpp=lambda l: 24 * l - 160,
)
newton.calculate([3])
def run(step, point_type, method, function):
# Validate function
if function == "rosenbrock":
n = 100
f = Rosenbrock(n)
elif function == "wood":
n = 4
f = Wood()
elif function == "mnist":
f = MNIST()
n = f.get_size()
else:
print("\n Error, invalid function")
quit()
# Initial point
if point_type == "const":
if function == "rosenbrock":
x0 = np.ones(n)
x0[0] = -1.2
x0[-2] = -1.2
elif function == "mnist":
x0 = np.ones(n)
else:
x0 = np.array([-3, -1, -3, -1])
elif point_type == "rand":
x0 = np.random.uniform(-1, 1, n)
else:
print("\n Error, invalid type of point")
quit()
mxitr = 50000 # Max number of iterations
tol_g = 1e-8 # Tolerance for gradient
tol_x = 1e-8 # Tolerance for x
if function == "mnist":
tol_f = 1e-3 # Tolerance for function
else:
tol_f = 1e-8 # Tolerance for function
# Method for step update
if step not in ["cubic", "barzilai", "zhang"]:
print("\n Error, invalid step method")
quit()
# Estimate minimum point through optimization method chosen
if method == "gd":
alg = GD()
elif method == "newton":
alg = Newton()
else:
print("\n Error: Invalid optimization method: %s\n" % method)
quit()
xs = alg.iterate(x0, mxitr, tol_g, tol_x, tol_f, f, step, function)
# Print point x found and function value f(x)
# print("\nPoint x found: ", xs[-1])
print("\nf(x) = ", f.eval(xs[-1]))
plt.plot(np.array(range(n)), xs[-1])
plt.legend(['x*'], loc='best')
plt.show()
if function == "mnist":
im = xs[-1][:-1].reshape(28, -1)
plt.imshow(im, cmap=plt.cm.gray)
plt.show()
from fractal import Fractal
from julia import Julia
from mandelbrot import Mandelbrot
from newton import Newton
from pheonix import Pheonix
# Static constant fractal instances.
MANDELBROT = Mandelbrot("Mandelbrot Set")
NEWTON = Newton("Newton Fractal")
JULIA = Julia("Julia Set")
PHEONIX = Pheonix("Pheonix Fractal")
def getFractal(fractal_id):
return Fractal.FRACTALS[fractal_id]
def getFractals():
return Fractal.FRACTALS
print("Formalizing function by tensorflow...")
sess = tf.Session()
if args.methods[0] == "1":
print("Option 1: Applying Bisection technique...")
model = Bisection(sess, func, x_tensorflow,
DEFAULT_SETTINGS["lbound"],
DEFAULT_SETTINGS["rbound"],
DEFAULT_SETTINGS["step"],
DEFAULT_SETTINGS["floating_point"])
model.find_valid_intervals()
model.find_roots()
elif args.methods[0] == "2":
print("Option 2: Applying Newton's technique...")
model = Newton(sess, func, x_tensorflow,
DEFAULT_SETTINGS["lbound"],
DEFAULT_SETTINGS["rbound"],
DEFAULT_SETTINGS["step"],
DEFAULT_SETTINGS["floating_point"],
DEFAULT_SETTINGS["iterator"])
model.find_valid_intervals()
model.find_roots()
if len(model.roots) > 0:
print("Approximately root of the function are:")
for iter, i in enumerate(model.roots):
print("root %d:%f" % (iter, i))
print()
print("#" * 20)
print("Press q to quit, or any button to start")
key = input()