def PSO_purana(costFunc, bounds, maxiter, swarm_init=None):
num_dimensions = len(swarm_init[0])
err_best_g = -1 # best error for group
pos_best_g = [] # best position for group
num_particles = len(swarm_init)
# establish the swarm
swarm = create_n_particles(num_particles, num_dimensions, swarm_init)
# begin optimization loop
i = 0
while i < maxiter:
#print i,err_best_g
# cycle through particles in swarm and evaluate fitness
for j in range(0, num_particles):
swarm[j]['pos_best_i'], swarm[j]['err_best_i'] = evaluate(
costFunc, swarm[j])
# determine if current particle is the best (globally)
if swarm[j]['err_i'] < err_best_g or err_best_g == -1:
pos_best_g = list(swarm[j]['position_i'])
err_best_g = float(swarm[j]['err_i'])
# cycle through swarm and update velocities and position
for j in range(0, num_particles):
swarm[j]['velocity_i'] = update_velocity(pos_best_g, swarm[j])
swarm[j]['position_i'] = update_position(bounds, swarm[j])
i += 1
# print final results
#print ('\n')
#print (pos_best_g,' , ', err_best_g)
return pos_best_g[0], err_best_g
def PSO(costFunc,
bounds,
maxiter,
shared_list,
return_list,
l,
num_particles=None,
swarm_init=None,
log=True,
t_list=None):
num_dimensions = len(swarm_init[0])
err_best_g = -1 # best error for group
pos_best_g = [] # best position for group
num_particles = len(swarm_init)
# establish the swarm
swarm = create_n_particles(num_particles, num_dimensions, swarm_init)
# begin optimization loop
i = 0
while i < maxiter:
#print i,err_best_g
# cycle through particles in swarm and evaluate fitness
for j in range(0, num_particles):
best_pos, swarm[j]['err_best_i'] = evaluate(costFunc, swarm[j])
swarm[j]['pos_best_i'] = best_pos
# determine if current particle is the best (globally)
if swarm[j]['err_i'] < err_best_g or err_best_g == -1:
pos_best_g = list(swarm[j]['position_i'])
err_best_g = float(swarm[j]['err_i'])
# update the global best in the manager list after k iterations
# we need to add some mutex lock here
if i == maxiter // 2:
l.acquire()
best_galactic_pos = shared_list[0]
best_galactic_err = shared_list[1]
#print("best_galactic_err: " ,best_galactic_err)
#print("best_galactic_pos: ", best_galactic_pos)
if err_best_g < best_galactic_err and err_best_g != -1:
shared_list[1] = err_best_g
#print(err_best_g)
shared_list[0] = pos_best_g
else:
#print("changing pos_best_g from", pos_best_g, " to ", best_galactic_pos)
#emp_list = []
err_best_g = float(best_galactic_err)
#emp_list.append(best_galactic_pos)
pos_best_g = [best_galactic_pos]
l.release()
# cycle through swarm and update velocities and position
for j in range(0, num_particles):
swarm[j]['velocity_i'] = update_velocity(pos_best_g, swarm[j])
swarm[j]['position_i'] = update_position(bounds, swarm[j])
if log:
t_list.append(err_best_g)
i += 1
return_list.append(pos_best_g[0])
def PSO(classifier,
bounds,
maxiter,
shared_list,
return_list,
l,
num_particles=None,
swarm_init=None,
pso_train_data=None):
# create minibatches inside PSO
num_dimensions = len(swarm_init[0])
err_best_g = -1 # best error for group
pos_best_g = [] # best position for group
num_particles = len(swarm_init)
#print('adress of classifier object is: ', id(classifier))
# establish the swarm
# initialize swarm population
#print('len(swarm_init): ', len(swarm_init), 'shape of swarm_init[0]: ', swarm_init[0].shape, '\n')
swarm = create_n_particles(num_particles, num_dimensions, swarm_init)
# begin optimization loop
i = 0
while i < maxiter:
#print i,err_best_g
# cycle through particles in swarm and evaluate fitness
for j in range(0, num_particles):
best_pos, swarm[j]['err_best_i'] = evaluate(
classifier, swarm[j], pso_train_data)
swarm[j]['pos_best_i'] = best_pos
# determine if current particle is the best (globally)
if swarm[j]['err_i'] < err_best_g or err_best_g == -1:
pos_best_g = list(swarm[j]['position_i'])
err_best_g = float(swarm[j]['err_i'])
# update the global best in the manager list after k iterations
# we need to add some mutex lock here
if i == maxiter // 2:
l.acquire()
best_galactic_pos = shared_list[0]
best_galactic_err = shared_list[1]
#print("best_galactic_err: " ,best_galactic_err)
#print("best_galactic_pos: ", best_galactic_pos)
if err_best_g < best_galactic_err:
shared_list[1] = err_best_g
#print(err_best_g)
shared_list[0] = pos_best_g
else:
#print("changing pos_best_g from", pos_best_g, " to ", best_galactic_pos)
#emp_list = []
err_best_g = float(best_galactic_err)
#emp_list.append(best_galactic_pos)
pos_best_g = [best_galactic_pos]
l.release()
# cycle through swarm and update velocities and position
for j in range(0, num_particles):
swarm[j]['velocity_i'] = update_velocity(pos_best_g, swarm[j])
swarm[j]['position_i'] = update_position(bounds, swarm[j])
i += 1
#print('shape of swarm[0][position_i] is: ', swarm[0]['position_i'].shape)
return_list.append((pos_best_g[0], swarm[:]['position_i']))