def callback(event, monitor):
callbacks.neat_callback(event, monitor)
global epochcount
global mode
if event == 'epoch.end':
#gc.collect()
maxfitness = monitor.evoTask.curSession.pop.inds[0]['fitness']
maxfitness_ind = monitor.evoTask.curSession.pop.inds[0]
maxfitness_records.append(maxfitness)
# 如果最大适应度达到了env.max_notdone_count（说明对当前环境已经产生适应），或者进化迭代数超过10次（当前环境下适应达到最大）
# 则提升复杂度
epochcount += 1
if maxfitness >= env.max_notdone_count or epochcount >= maxepochcount:
#保存复杂度和对应最大适应度记录
fitness_records.append(maxfitness)
complex_records.append(force.force_generator.currentComplex())
print([(f, c) for f, c in zip(complex_records, fitness_records)])
np.save('hyperneat_result', (complex_records, fitness_records))
# 保存过程中每一步的最大适应度记录
filename = os.path.split(os.path.realpath(__file__))[0] + os.sep + 'datas_' + mode + os.sep + \
'hyperneat' + os.sep + 'hyperneat.csv'
out = open(filename, 'a', newline='')
csv_write = csv.writer(out, dialect='excel')
csv_write.writerow([complex_records[-1]] + maxfitness_records)
maxfitness_records.clear()
# 保存最优基因网络拓扑
filename = os.path.split(os.path.realpath(__file__))[0] + os.sep + 'datas_' + mode + os.sep + \
'hyperneat'+os.sep + 'hyperneat_' + str(complex_records[-1]) + '_ind' + str(maxfitness_ind.id) + '_' + \
str(maxfitness) + '.svg'
netviewer = NetworkView()
netviewer.drawNet(maxfitness_ind.genome,
filename=filename,
view=False)
# 提升复杂度
changed, maxcomplex, k, w, f, sigma = force.force_generator.promptComplex(
complexunit)
if changed and maxcomplex is not None:
print('环境复杂度=%.3f,k=%.2f,w=%.2f,f=%.2f,sigma=%.2f' %
(maxcomplex, k, w, f, sigma))
if mode == 'reset':
monitor.evoTask.curSession.runParam.terminated.maxIterCount = epochcount - 1
else:
np.save('hyperneat_result', complex_records, fitness_records)
#plt.plot(complex_records, reward_list, label='reward')
plt.plot(complex_records, fitness_records, label='times')
plt.xlabel('complexes')
plt.savefig('./hyperneat_cartpole.png')
epochcount = 0
elif event == 'session.end':
filename = 'singlecartpole.session.' + str(
monitor.evoTask.curSession.taskxh) + '.mov'
eliest = monitor.evoTask.curSession.pop.eliest
def neat_callback(event, monitor):
# session结束的时候将精英个体的图形写入文件
if event == 'session.end':
filename = 'session' + str(monitor.evoTask.curSession.taskxh) + ".ind"
eliest = monitor.evoTask.curSession.pop.eliest
if collections.isEmpty(eliest): return
optimaInd = eliest[0]
net = optimaInd.genome
netviewer = NetworkView()
netviewer.drawNet(net,
filename=filename + str(optimaInd.id) + '.svg',
view=False)
#collections.foreach(eliest,lambda ind : netviewer.drawNet(filename=filename+ind.id+'.svg',view=False))
#evolutionView = EvolutionViewer()
#evolutionView.drawSession(monitor,monitor.evoTask.curSession,'fitness')
return True
def callback(event, monitor):
callbacks.neat_callback(event, monitor)
global epochcount
global mode
global exec_xh
global epoch_maxfitness
global epoch_maxfitness_ind
if event == 'epoch.end':
gc.collect()
maxfitness = monitor.evoTask.curSession.pop.inds[0]['fitness']
maxfitness_ind = monitor.evoTask.curSession.pop.inds[0]
maxfitness_records.append(maxfitness)
if maxfitness > epoch_maxfitness:
epoch_maxfitness = maxfitness
epoch_maxfitness_ind = maxfitness_ind
# 如果最大适应度达到了env.max_notdone_count（说明对当前环境已经产生适应），或者进化迭代数超过10次（当前环境下适应达到最大）
# 则提升复杂度
epochcount += 1
if maxfitness >= env.max_notdone_count or epochcount >= maxepochcount:
# 保存复杂度和对应最大适应度记录
fitness_records.append(epoch_maxfitness)
complex_records.append(force.force_generator.currentComplex())
print([(f, c) for f, c in zip(complex_records, fitness_records)])
#np.save('neat_result', (complex_records, fitness_records))
# 保存过程中每一步的最大适应度记录
filename = neatdatapath + 'neat_' + mode + '_' + str(
exec_xh) + '.csv'
out = open(filename, 'a', newline='')
csv_write = csv.writer(out, dialect='excel')
csv_write.writerow([complex_records[-1]] + maxfitness_records)
maxfitness_records.clear()
# 保存最优基因网络拓扑
filename = neatdatapath + 'neat_' + mode + '_' + str(exec_xh) + '_' + \
str(complex_records[-1]) + '_' + str(epoch_maxfitness_ind.id) + '_' + \
str(epoch_maxfitness) + '.svg'
netviewer = NetworkView()
netviewer.drawNet(epoch_maxfitness_ind.genome,
filename=filename,
view=False)
# 计算最优网络模块度
modularity = epoch_maxfitness_ind.genome.compute_modular()
value, parts = modularity
modularities.append((complex_records[-1], modularity))
print(modularities)
filename = neatdatapath + 'neat_' + mode + '_' + str(
exec_xh) + '_' + 'modularity.csv'
out = open(filename, 'a', newline='')
csv_write = csv.writer(out, dialect='excel')
csv_write.writerow([complex_records[-1], value, parts])
# 提升复杂度
changed, maxcomplex, k, w, f, sigma = force.force_generator.promptComplex(
complexunit)
if changed:
print('环境复杂度=%.3f,k=%.2f,w=%.2f,f=%.2f,sigma=%.2f' %
(maxcomplex, k, w, f, sigma))
if mode == 'reset':
monitor.evoTask.curSession.runParam.terminated.maxIterCount = epochcount - 1
else:
return
epoch_maxfitness = 0.
epoch_maxfitness_ind = None
epochcount = 0
elif event == 'session.end':
filename = 'singlecartpole.session.' + str(
monitor.evoTask.curSession.taskxh) + '.mov'
eliest = monitor.evoTask.curSession.pop.eliest
def callback(event, monitor):
callbacks.neat_callback(event, monitor)
global cur_state_maxfitness
global cur_state_maxfitness_ind
global cur_state_maxfitnesses_per_epoch
global cur_state_no
global epoch_no
global mode
if event == 'epoch.end':
gc.collect()
epoch_no += 1
# 记录这轮的最大适应度
maxfitness = monitor.evoTask.curSession.pop.inds[0]['fitness']
maxfitness_ind = monitor.evoTask.curSession.pop.inds[0]
if maxfitness > cur_state_maxfitness:
cur_state_maxfitness = maxfitness
cur_state_maxfitness_ind = maxfitness_ind
cur_state_maxfitnesses_per_epoch.append(maxfitness)
# 如果最大适应度达到了env.max_notdone_count（说明对当前环境已经产生适应），或者进化迭代数超过10次（当前环境下适应达到最大）
# 则提升复杂度
if maxfitness >= MAX_FITNESS or epoch_no >= MAX_STATE_EPOCH:
# 保存复杂度和对应最大适应度记录
state_records.append((cur_state_maxfitnesses_per_epoch,
preyController.current_complex))
print([(max(r[0]), r[1]) for r in state_records])
# 保存过程中每一步的最大适应度记录
filename = log_path + os.sep + 'neat_' + mode + '.csv'
out = open(filename, 'a', newline='')
csv_write = csv.writer(out, dialect='excel')
csv_write.writerow(state_records[-1])
# 保存最优基因网络拓扑
try:
filename = log_path + os.sep + 'neat_' + mode + '_' + str(cur_state_no) + '_' + str(
maxfitness_ind.id) + '_' + \
str(maxfitness) + '.svg'
netviewer = NetworkView()
netviewer.drawNet(maxfitness_ind.genome,
filename=filename,
view=False)
except RuntimeError as err:
pass
# 清空当前状态
cur_state_no += 1
cur_state_maxfitness = 0.
cur_state_maxfitness_ind = None
cur_state_maxfitnesses_per_epoch = []
epoch_no
# 提升复杂度
changed = preyController.promote()
if changed:
print('复杂度提升,state_no=', str(cur_state_no), ',complex=',
preyController.current_complex)
else:
maxfitness_records = [max(s[0]) for s in state_records]
complex_records = [s[1] for s in state_records]
plt.plot(complex_records, maxfitness_records, label='times')
plt.xlabel('complexes')
filename = log_path + os.sep + 'neat_' + mode + '.png'
plt.savefig(filename)
return
elif event == 'session.end':
pass