def afpo(self):
envs = ENVIRONMENTS()
print('envs to run:', c.envsToRun, ' pop init:',
c.num_rand_phc + len(self.fit_ids))
parents = POPULATION(c.num_rand_phc + len(self.fit_ids))
parents.InitializeRandomPop()
# parents = self.phc_ga(parents, envs, c.afpo2_phc_gens)
# print('\n############# phc done #############\n')
parents.Evaluate(envs, False, True)
for i in range(0, len(parents.p)):
parents.p[i].age += i
# for i in range(0, len(parents.p)):
# print('p[' + str(i) + '].age =', parents.p[i].age)
print("Initial population age adjusted")
parents.Print()
finalpop = self.af_pareto_ga(parents, envs)
frontinds = self.af_pareto_math(finalpop.GetAges(), finalpop.GetFits())
finalpopfront = POPULATION(len(frontinds))
for i in range(0, len(frontinds)):
finalpopfront.p[i] = finalpop.p[frontinds[i]]
finalpopfront.Pickling618()
self.ShowEvaluatePop(finalpopfront, envs, pp=True, pb=False)
def simple_ga(self, envs):
def inner(parents, envs, i):
# children = POPULATION(len(parents.p))
children = copy.deepcopy(parents)
children.FillFrom(parents)
children.Evaluate(envs, False, True)
print("\n*** Generation", i, "***")
print("\n*** Parents ***")
parents.Print()
print("*** Children ***")
children.Print()
print()
parents = children
return parents
envs = ENVIRONMENTS()
parents = POPULATION(c.popsize_simple_ga)
print('envs:', c.envsToRun, ' pop init:', c.popsize_simple_ga)
parents.InitializeRandomPop()
parents.Evaluate(envs, False, True)
# if c.gen_mode == 'time':
# t_end = time.time() + 60 * c.minutes
# i = 0
# while(time.time() < t_end):
# parents = inner(parents, envs, i)
# i += 1
# elif c.gen_mode == 'gens':
# for i in range(0, c.gens):
# parents = inner(parents, envs, i)
for i in range(0, c.numGensDivPop):
parents = inner(parents, envs, i)
return parents.p[0]
def af_pareto_spop(self, front, envs):
target_popsize = 20
pcc = POPULATION(target_popsize)
single = POPULATION(1)
single.InitializeRandomPop()
single.Evaluate(envs, False, True)
num_remain = target_popsize - (len(front.p) + 1)
if num_remain > 0:
pccr = POPULATION(num_remain)
for i in range(0, num_remain):
pccr.p[i] = copy.deepcopy(pcc.p[i % len(front.p)])
pccr.Mutate()
pccr.Evaluate()
for i in range(0, target_popsize):
if i == 0:
pcc.p[i] = single.p[0]
elif i < len(front.p) + 1:
pcc.p[i] = front.p[i - 1]
else:
pcc.p[i] = pccr[i - (len(front) + 1)]
ages = pcc.GetAges()
fits = pcc.GetFits()
print('fits', fits, ' ages', ages)
fits_inv = []
for i in range(0, len(fits)):
if fits[i] != 0:
fits_inv.append(fits[i]**-1)
else:
fits_inv.append(10**6)
dominant_inds = np.array(self.af_pareto_math(ages, fits_inv))
new_pop = POPULATION(len(dominant_inds))
for i in range(0, target_popsize):
dominant_inds_index = i % len(dominant_inds)
new_pop.p[i] = copy.deepcopy(pcc.p[dominant_inds_index])
if i >= len(dominant_inds):
new_pop.p[i].Mutate()
del pcc
new_pop.Print()
return new_pop
def main(self):
envs = ENVIRONMENTS()
parents = POPULATION(c.num_rand_phc + len(self.fit_ids))
print('envs:', c.envsToRun, ' pop init:',
c.num_rand_phc + len(self.fit_ids))
parents.InitializeRandomPop()
parents.Evaluate(envs, False, True)
for i in range(0, len(parents.p)):
parents.p[i].age += i
self.afpo(parents, envs)
def af_pareto_spop(self, parents, envs):
target_popsize = c.target_afpop
# pcc new population with target popsize
pcc = POPULATION(target_popsize)
# pcc.InitializeRandomPop() # not necessary..?
# add new random individual
single = POPULATION(1)
single.InitializeRandomPop()
# print("parents print?")
# parents.Print()
pcc.p[0] = single.p[0]
for i in range(1, target_popsize):
# print("i?", i, " size of pop?", len(parents.p))
pcc.p[i] = copy.deepcopy(parents.p[i % len(parents.p)])
if i % 2 == 0:
pcc.p[i].Mutate()
pcc.Evaluate(envs, False, True)
ages = pcc.GetAges()
fits = pcc.GetFits()
print('fits', fits)
print('ages', ages)
# if fitness is 0 inverse fitness set to 10000
fits_inv = []
for i in range(0, len(fits)):
if fits[i] != 0:
fits_inv.append(fits[i]**-1)
else:
fits_inv.append(10**4)
dominant_inds = np.array(self.af_pareto_math(ages, fits_inv))
print("dominant inds", dominant_inds)
self.graph_both3(ages, fits_inv, dominant_inds)
# self.graph_both2(ages, fits_inv, dominant_inds) # dominant inds not right, should be list 1s and 0s
# self.graph_both(pareto_front_x, pareto_front_y, xx_copy, yy_copy, popsize, ymax)
new_pop = POPULATION(len(dominant_inds))
for i in range(0, target_popsize):
dominant_inds_index = i % len(dominant_inds)
new_pop.p[i] = copy.deepcopy(pcc.p[dominant_inds_index])
if i >= len(dominant_inds):
new_pop.p[i].Mutate()
del pcc
# new_pop.Print()
return new_pop
def main(self):
t_end = time.time() + 60 * c.minutes
envs = ENVIRONMENTS()
print('envs to run:', c.envsToRun, ' initial popsize:', c.num_rand_phc)
parents = POPULATION(c.num_rand_phc + len(self.fit_ids))
parents.InitializeRandomPop()
parents.Evaluate(envs, pp=False, pb=True)
parents.Print()
if c.alg == 'afpo2':
self.afpo()
elif c.alg == 'garef':
parents = self.garef(c.numGensDivPop, envs, parents, t_end)
self.ShowEvaluatePop(parents, envs)
def af_pareto_spop(self, parents, envs):
children = copy.deepcopy(parents)
children.Mutate()
single = POPULATION(1)
single.InitializeRandomPop()
single.Evaluate(envs, pp=False, pb=True)
parents.Evaluate(envs, pp=False, pb=True)
children.Evaluate(envs, pp=False, pb=True)
pcc = POPULATION(1 + 2 * len(parents.p))
pcc.p[0] = single.p[0]
for i in range(1, len(parents.p) + 1):
pcc.p[i] = parents.p[i - 1]
for i in range(
len(parents.p) + 1,
len(parents.p) + len(children.p) + 1):
pcc.p[i] = children.p[i - len(children.p) - 1]
s_fits = single.GetFits()
s_ages = single.GetAges()
p_fits = parents.GetFits()
p_ages = parents.GetAges()
c_fits = children.GetFits()
c_ages = children.GetAges()
pc_fits = np.concatenate((p_fits, c_fits))
pc_fits = np.concatenate((s_fits, pc_fits))
pc_ages = np.concatenate((p_ages, c_ages))
pc_ages = np.concatenate((s_ages, pc_ages))
pc_fits_trim = {}
pc_ages_trim = {}
pc_fits_nz_inds = []
for i in range(0, len(pc_fits)):
if pc_fits[i] != 0:
pc_fits_trim[i] = pc_fits[i]
pc_ages_trim[i] = pc_ages[i]
pc_fits_nz_inds.append(i)
if len(pc_fits_trim) == 0:
print('none fit')
exit(0)
pcc_trim = POPULATION(len(pc_fits_trim))
for i in range(0, len(pc_fits_trim)):
pcc_trim.p[i] = pcc.p[pc_fits_nz_inds[i]]
pc_fits_inv_trim = {}
print('pc_fits_trim', pc_fits_trim)
for i in pc_fits_trim.keys():
pc_fits_inv_trim[i] = pc_fits_trim[i]**-1
# find max val in pc_fits_inv_trim
maxval = 0
for i in pc_fits_inv_trim.values():
if i > maxval: maxval = i
# returns a list right now
dominant_inds = np.array(
self.af_pareto_math(pc_ages_trim, pc_fits_inv_trim, maxval * 1.05))
if c.print_func == 1: print('***** exit af_pareto_math *****\n')
print('dominant inds', dominant_inds)
# new_pop = POPULATION(len(dominant_inds))
# count = 0
print('******* len of pcc_trim.p', len(pcc_trim.p), '*******')
# for i in dominant_inds:
# print('i in dominant inds is', i)
# new_pop.p[count] = pcc_trim.p[i]
# count += 1
for i in range(0, c.target_afpop):
print('dinds!', dominant_inds[i % len(dominant_inds)])
new_pop = POPULATION(c.target_afpop)
for i in range(0, c.target_afpop):
new_pop.p[i] = copy.deepcopy(
pcc_trim.p[[dominant_inds[i % len(dominant_inds)]]])
print('new pop len', len(new_pop.p))
return new_pop
parents = POPULATION(c.popSize)
if k < num_rand:
first = np.random.random_sample((c.numInputs, c.numHidden)) * 2 - 1
second = np.random.random_sample((c.numHidden, c.numOutputs)) * 2 - 1
genome = [first, second]
else:
filename = 'genome' + str(fit_labels[0])
del fit_labels[0]
pathLoad = c.pathLoad
filename = pathLoad + filename
g = open(filename, 'r')
genome = pickle.load(g)
g.close()
# parents.InitializeUniformPop(genome)
parents.p[k] = INDIVIDUAL(k, genome, 0)
parents.Evaluate(envs, pp=False, pb=True)
parents.Print()
for i in range(0, c.numGensDivPop):
# the POPULATION constructor creates an empty dictionary and stores the variable popSize
children = POPULATION(c.popSize)
# children.Print('bm')
print()
children.FillFrom(parents)
print('seconds remaining', t_end - time.time())
# print('\n0', end=' ')
parents.Print()
children.Evaluate(envs, pp=False, pb=True)
# print(i, end=' ')
children.Print(i)
def af_pareto_spop(self, parents, envs):
children = copy.deepcopy(parents)
children.Mutate()
single = POPULATION(1)
single.InitializeRandomPop()
single.Evaluate(envs, pp=False, pb=True)
parents.Evaluate(envs, pp=False, pb=True)
children.Evaluate(envs, pp=False, pb=True)
# combine single, parents, children into one pop
pcc = POPULATION(1 + 2 * len(parents.p))
pcc.p[0] = single.p[0]
for i in range(1, len(parents.p) + 1):
pcc.p[i] = parents.p[i - 1]
for i in range(len(parents.p) + 1, len(parents.p) + len(children.p) + 1):
pcc.p[i] = children.p[i - len(children.p) - 1]
# these methods in pop class all return lists
s_fits = single.GetFits()
s_ages = single.GetAges()
p_fits = parents.GetFits()
p_ages = parents.GetAges()
c_fits = children.GetFits()
c_ages = children.GetAges()
pc_fits = np.concatenate((p_fits, c_fits))
pc_fits = np.concatenate((s_fits, pc_fits))
pc_ages = np.concatenate((p_ages, c_ages))
pc_ages = np.concatenate((s_ages, pc_ages))
min_fitness_pc = self.return_second_highest_in_list_or_dict(pc_fits)
for i in range(0, len(pc_fits)):
if pc_fits[i] == 0:
pc_fits[i] = min_fitness_pc / 10.0
pcc.p[i].fitness = min_fitness_pc / 10.0
pc_fits_inv = []
print('pc_fits', pc_fits)
for i in range(0, len(pc_fits)):
pc_fits_inv.append(pc_fits[i]**-1)
# returns a list right now
dominant_inds = np.array(self.af_pareto_math(pc_ages, pc_fits_inv))
if c.print_func == 1:
print('********** exit af_pareto_math **********\n')
print('dominant inds', dominant_inds)
# new_pop = POPULATION(len(dominant_inds))
# count = 0
print('******* len of pcc.p', len(pcc.p), '*******')
# for i in dominant_inds:
# print('i in dominant inds is', i)
# new_pop.p[count] = pcc_trim.p[i]
# count += 1
if len(dominant_inds) < c.target_afpop:
new_popsize = c.target_afpop
else:
new_popsize = len(dominant_inds)
print("new pop size ", new_popsize)
new_pop = POPULATION(new_popsize)
for i in range(0, new_popsize):
# print("type of pcc.p[dominant_inds[" + str(i % len(dominant_inds)) + "]]", type(pcc.p[dominant_inds[i % len(dominant_inds)]]))
new_pop.p[i] = copy.deepcopy( pcc.p[dominant_inds[i % len(dominant_inds)]] )
# print(new_pop.p[i].age)
print('new pop len', len(new_pop.p))
return new_pop
from pop import POPULATION
from afpo5t2 import GENETIC_ALGORITHM
import const as c
import time
import pyrosim
import matplotlib.pyplot as plt
import numpy as np
import random
import copy
import pickle
import os
alg = GENETIC_ALGORITHM()
mode = 't'
mode = 'r'
if mode == 't':
envs = ENVIRONMENTS()
pop = POPULATION(10)
pop.InitializeRandomPop()
pop.Evaluate(envs, False, True)
for i in range(0, len(pop.p)):
pop.p[i].age += i
pop.Print()
frontpop = alg.gen_par_front_pop(pop, envs)
frontpop.Print()
else:
# alg.main()
alg.combo_simple_phc_ga()
filename = 'genome' + str(fit_labels[0])
del fit_labels[0]
pathLoad = c.pathLoad
filename = pathLoad + filename
g = open(filename, 'r')
genome = pickle.load(g)
g.close()
parentsF.p[k] = INDIVIDUAL(k, genome)
for e in envs.envs:
parentsF.p[k].Start_Evaluation(envs.envs[e],
pp=False,
pb=True,
send=True)
parentsF.Evaluate(envs, pp=False, pb=True)
parentsF.Print('population size', k)
print('################ Begin parallel hill climber ######################')
while time.time() < t_end:
# print('############# i', i)
print('seconds remaining', t_end - time.time())
print('\nF', end='')
parentsF.Print()
childrenF = copy.deepcopy(parentsF)
childrenF.Mutate()
childrenF.Evaluate(envs, pp=False, pb=True)
print('F', end='')
childrenF.Print(i)
parentsF.ReplaceWith(childrenF)