def __init__(self, N, L, rho, eta):
"""Creates and manages the field, which is not a lattice, but a continuous
square plane. Only two parameters are needed; set the other to zero. Assumes a
radius around the particle of 1.
N
Number of particles to create.
L
Side length of the square field. Must be an integer.
rho
Density, approximately equivalent to N/L^2.
"""
# Handle the choose-any-two nature of the parameters
self.N = (N if (N != 0 and N is not None) else int(rho * L * L))
self.width = (L if
(L != 0 and L is not None) else int(math.sqrt(N / rho)))
self.height = self.width
self.eta = eta
# generate the N agents, stored in a triple-nested list. Use this to
# iterate over all particles in a given cell:
# do_stuff(p) for p in self.particles[y][x]
self.particles = [[[] for __ in range(self.width)]
for _ in range(self.height)]
x = y = 0
for i in range(self.N):
x = self.width * U()
y = self.height * U()
self.particles[int(y)][int(x)].append(Particle(x, y))
def naiv_parametrix(T, Lambda, counter):
#naiv implementation of the forward method for a simple exemple
X_pred = x0
TETA = 1
#
t = 0
delta_tau = -(1. / Lambda) * log(U(a=0, b=1))
t += delta_tau
Var = 0
if (delta_tau > T):
counter = counter + 1
X_new = x0 + b(x0) * T + sig(x0) * sqrt(T) * W(0, 1)
Var = (exp(Lambda * T) * f(X_new))**2
return exp(Lambda * T) * f(X_new), counter, Var
else:
while (t < T):
X_new = X_pred + delta_tau * b(X_pred) + sqrt(delta_tau) * W(
0, 1) * sig(X_pred)
TETA = TETA * teta(delta_tau, X_pred, X_new) / Lambda
X_pred = X_new
delta_tau = -(1. / Lambda) * log(U(a=0, b=1))
t = t + delta_tau
delta = T - t + delta_tau
last_X = X_pred + delta * b(X_pred) + sqrt(delta) * W(0,
1) * sig(X_pred)
TETA = TETA * teta(delta, X_pred, last_X) / Lambda
Var = (exp(Lambda * T) * f(last_X) * TETA)**2
return exp(Lambda * T) * f(last_X) * TETA, counter, Var
def parametrix_plot(T, Lambda, counter):
#modification de naive_parmetrix pour l'affichage
X_pred = x0
TETA = 1
t = 0
delta_tau = -(1. / Lambda) * log(U(a=0, b=1))
t += delta_tau
Var = 0
bool = True
if (delta_tau > T):
counter = counter + 1
X_new = x0 + b(x0) * T + sig(x0) * sqrt(T) * W(0, 1)
Var = (exp(Lambda * T) * f(X_new))**2
return exp(Lambda * T) * f(X_new), Var, counter, bool, X_new
else:
while (t < T):
X_new = X_pred + delta_tau * b(X_pred) + sqrt(delta_tau) * W(
0, 1) * sig(X_pred)
TETA = TETA * teta(delta_tau, X_pred, X_new) / Lambda
X_pred = X_new
delta_tau = -(1. / Lambda) * log(U(a=0, b=1))
t = t + delta_tau
delta = T - t + delta_tau
last_X = X_pred + delta * b(X_pred) + sqrt(delta) * W(0,
1) * sig(X_pred)
TETA = TETA * teta(delta, X_pred, last_X) / Lambda
Var = (exp(Lambda * T) * f(last_X) * TETA)**2
bool = False
return exp(Lambda * T) * f(last_X) * TETA, Var, counter, bool, last_X
def _optimization_step(self, param):
"""
This method is run on a separate process at each iteration.
Applies the formula (*) - see it below - for one particle
:param param: contains particle, velocity, pbest and gbest, all needed to implement formula
:return: returns updated values for particle, velocity, pbest and gbest
"""
particle, velocity, pbest, gbest = param
# save particle because we will compute velocity at the end
old_particle = deepcopy(particle)
old_particle = op_perm_fix(x=self.full_particle(old_particle),
P=self.precedences)
# apply formula (*): v(k+1) = [inertia * v(k)] + [personal * rand() * (p(i) - x(i))] + [social * rand() * (g - x(i))]
# compute each term inside square brackets
velocity_inertia = op_scalar_mul_velocity(c=self.coef_inertia,
v=velocity)
diff_velocity_personal = op_perm_sub_perm(pbest, particle)
velocity_personal = op_scalar_mul_velocity(
self.coef_personal * U(0, 1), diff_velocity_personal)
diff_velocity_social = op_perm_sub_perm(gbest, particle)
velocity_social = op_scalar_mul_velocity(self.coef_social * U(0, 1),
diff_velocity_social)
# apply each velocity individually to particle because we don't have a velocity + velocity operator
particle = op_perm_sum_velocity(particle, velocity_inertia)
particle = op_perm_sum_velocity(particle, velocity_personal)
particle = op_perm_sum_velocity(particle, velocity_social)
# fix current particle so that it repects precedence constraints
particle = op_perm_fix(x=self.full_particle(particle),
P=self.precedences)
# compute velocity that transforms old_particle into self.patricles[i]
velocity = op_perm_sub_perm(particle, old_particle)
cost = self.cost(particle)
# update personal best in case we got a better particle than previous personal best
if cost < self.cost(pbest):
pbest = deepcopy(particle)
if cost < self.cost(self.gbest):
self.gbest = deepcopy(particle)
return particle, velocity, pbest, cost
def update(self):
"""Generates a new list of particles, computes their velocities, and then does
an integration step (delta t = 1) to determine the new particle's x and y position.
"""
new_particles = [[[] for __ in range(self.width)]
for _ in range(self.height)]
x = 0
y = 0
while y < self.height:
x = 0
while x < self.width:
for p in self.particles[y][x]: # p is each current particle
# updating velocity of new particle
new_angle = self.getLocalAverageAngle(
p, x, y) + self.eta * (U() - 0.5)
P = Particle(0, 0)
P.setVelocityAngleMagnitude(new_angle, 0.03)
# numerical integration with dt=1, and we are on a torus
P.xpos = (p.xpos + P.xvel) % self.width
P.ypos = (p.ypos + P.yvel) % self.height
# store in correct tile
new_particles[int(P.ypos)][int(P.xpos)].append(P)
x += 1
y += 1
# when finished updating everything, save the new agents and let the GC
# handle the old
self.particles = new_particles
def reject_gamma(a):
u = U(0,1)
if u < (exp(1)/(a+exp(1))):
Y = (((a+exp(1))/exp(1))*u)**(1./a)
else:
Y = -log((1-u)*((a+exp(1))/a*exp(1)))
if Y < 1:
q = exp(-Y)
if Y> 1 :
q = Y**(a-1)
else :
q = 0
v = U(0,1)
if v <= q:
return Y
else:
return reject_gamma(a)
def __init__(self, x, y, v=0.03): """This represents a single particle in the system. It is created with a velocity of ``v`` in a random direction, at point (x,y). """ self.xpos = x self.ypos = y theta = 2 * math.pi * U() # pick a random angle self.xvel = v * math.cos(theta) self.yvel = v * math.sin(theta)
def __init__(self, N, L, rho, eta): """Creates and manages the field, which is not a lattice, but a continuous square plane. Only two parameters are needed; set the other to zero. N Number of particles to create. L Side length of the square field. rho Density, equivalent to N/L^2. """ # Handle the choose-any-two nature of the parameters self.N = (N if N != 0 and N is not None else rho*L*L) self.width = (L if L != 0 and L is not None else int(math.sqrt(N/rho))) self.height = self.width self.eta = eta # generate the N agents self.particles = [Particle(self.width * U(), self.height * U()) for _ in range(self.N)]
def parametrix_beta(T, gamma, counter, tau_bar):
#naiv implementation of the forward method for a simple exemple
X_pred = x0
TETA = 1
t = 0
E = -1 * log(U(a=0, b=1))
delta_tau = tau_bar * exp(-(1 - gamma) * E)
t += delta_tau
Var = 0
Sn = np.array([t])
if (delta_tau > T):
counter = counter + 1
X_new = x0 + b(x0) * T + sig(x0) * sqrt(T) * W(0, 1)
Sn = np.array([0])
P = pn(tau_bar, Sn, T, gamma)
Var = (f(X_new) / P)**2
return f(X_new) / P, Var, counter
else:
while (t < T):
X_new = X_pred + delta_tau * b(X_pred) + sqrt(delta_tau) * W(
0, 1) * sig(X_pred)
TETA = TETA * teta(delta_tau, X_pred, X_new)
X_pred = X_new
E = -1 * log(U(a=0, b=1))
delta_tau = tau_bar * exp(-(1 - gamma) * E)
t = t + delta_tau
if (t < T):
Sn = np.append(Sn, t)
delta = T - t + delta_tau
last_X = X_pred + delta * b(X_pred) + sqrt(delta) * W(0,
1) * sig(X_pred)
TETA = TETA * teta(delta, X_pred, last_X)
P = pn(tau_bar, Sn, T, gamma)
Var = (f(last_X) * TETA / P)**2
return f(last_X) * TETA / P, Var, counter
def update(self, delta_t): """Generates a new list of particles, computes their velocities, and then does an integration step to determine the new particle's x and y position. """ new_particles = [Particle(0, 0) for _ in range(self.N)] # loop over both lists at once: for (p, P) in zip(self.particles, new_particles): # p is each current particle # P is each new particle neighbors = self.getNeighbors(p, 1) # updating velocity of new particle new_angle = math.atan2(sum(n.yvel for n in neighbors), sum(n.xvel for n in neighbors)) + self.eta*(U()-0.5) P.setVelocityAngleMagnitude(new_angle, 0.03) # numerical integration P.xpos = (p.xpos + P.xvel*delta_t) % self.width P.ypos = (p.ypos + P.yvel*delta_t) % self.height # when finished updating everything, save the new agents and let the GC # handle the old self.particles = new_particles
H = P(f)
if H.lower() == g:
if D:
A(h)
for S in D:
A(O(S, i))
break
I = j
for J in G:
if H in J.lower():
I = T
E.append(J)
if not I: A(k)
else:
K = [A for A in E for B in C if B in A.lower()]
if K: B = U(K)
else: B = U(E)
A(l.format(O(B, m)))
V.copy(B)
A(n)
L = P(X)
if o in L:
A(p)
D.append(B)
G.remove(B)
if not L:
for M in B:
if M.lower() in C: C.remove(M.lower())
elif W.lower() == 'french':
with Y('touslesmots.txt', b) as R:
G = R.read().splitlines()
def gauss(m,s):
u = U(a=0,b=1)
return inv_gauss(u)*sqrt(s)+m
def cauchy(L):
u = U(a=0,b=1)
return L*tan(pi*(u-0.5))
def expn(L):
u = U(a=0,b=1)
return -(1./L)*log(u)
def bern(p=0.5):
u = U(a=0,b=1)
if u < p:
return 1
else:
return 0
def gauss_Box_Muller(m,s):
R = sqrt(expn(0.5))
u = U(a=0,b=2*pi)
return np.array([(R* cos(u))*sqrt(s)+m,(R* sin(u))*sqrt(s)+m])
