def run():
# ehnal网络初始化
senal.ehnal_init()
netdef = brain.createNetDef(neuronCounts=[30, 6])
pop_param = evolution.createPopParam(size=50)
run_param = evolution.createRunParam()
session = Session()
monitor = session.run()
from pybullet_envs.bullet import bullet_client
from evolution.session import EvolutionTask
import ne.neat as neat
import ne.neat.config as config
import evolution
import domains.antbullet.env as env
import domains.antbullet.measure as measure
from brain.networks import NeuralNetworkTask
if __name__ == '__main__':
# NEAT 前馈网络
neat.neat_init()
task = NeuralNetworkTask()
netdef = config.createNetDef(task, neuronCounts=[30, 8])
popParam = evolution.createPopParam(indTypeName='network',
size=100,
elitistSize=0.01,
genomeDefinition=netdef,
evaluators=[
(env.fitness, 0.5, 'fitness'),
(measure.novelty, 0.5, 'novelty')
],
species=config.defaultSpeciesMethod())
runParam = evolution.createRunParam(
debug=True, operations="NSGA2,neat_crossmate,neat_mutate")
evolutionTask = EvolutionTask(1, popParam, env.callback)
evolutionTask.execute(runParam)
elif oper == 1:v += 1
else: v -= 1
newg = g
newg[index] = str(v)
return newg
def phenome_mutate(p):
g,pt = phenome_spilt(p)
newg = genome_mutate(g)
postfix = np.random.uniform(low=0, high=99)
if len(postfix) < 2: postfix = '0' + str(postfix)
return newg + postfix
#endregion
#region init
popParam = evolution.createPopParam(size=10000,evaluators=fitnessLandscape.fitness,genomeFactory=VEEGenomeFactory())
runParam = evolution.createRunParam(maxIterCount=1000,operations='')
generation,max_generation = 0,10000
pop = Population.create(popParam)
class Node:
def __init__(self,genome,level,inds,childs,parent):
self.genome = genome
self.level = level
self.inds = inds
self.childs = childs
self.parent = parent
self.features = {}
root = Node(init_genome,0,pop.inds,[],None)
'attributes': {}
}]
netdef.box = {
'activation_threadshold': 0.5, # 激活门限，小于该值相当于没有激活
'overflow_count': 20, # 溢出数量，超过该数量将导致节点分裂
'adjust_count': 10, # 调整数量，超出该数量将导致节点均值和协方差调整
'energy': [0., 1.], # 瞬时能量限制范围
'init_lambda': 1., # 初始协方差矩阵系数
'grid_size': 100, # 特征空间中每个维度的网格数，总网格数为维度个gird_size相乘
'benchmark_energy': 1.0 # 瞬时基准能耗
}
#endregion
#region 种群参数和运行参数
popParam = evolution.createPopParam(size=100,
elitistSize=0.05,
genomeDefinition=netdef)
runParam = evolution.createRunParam(
1000,
1.0,
mutate_propotion=0.2,
activity={
'stability_threshold': 0.8,
'stability_resdual': 0.2,
'stability_max_count': 3,
'stability_output_count': 3,
'autonomous_targe': 10
},
handlers={
'ind.begin': handle_ind_begin, # 个体活动初始化
'ind.action': handle_ind_action, # 个体输出动作
currentdir = os.path.dirname(os.path.abspath(inspect.getfile(inspect.currentframe())))
parentdir = os.path.dirname(os.path.dirname(currentdir))
os.sys.path.insert(0, parentdir)
from pybullet_envs.bullet import bullet_client
from evolution.session import EvolutionTask
import ne.neat as neat
import ne.neat.config as config
import evolution
import domains.antbullet.env as env
from brain.networks import NeuralNetworkTask
if __name__ == '__main__':
# NEAT 前馈网络
neat.neat_init()
task = NeuralNetworkTask()
netdef = config.createNetDef(task,neuronCounts=[30,8])
popParam = evolution.createPopParam(indTypeName='network',
size=100,
elitistSize=0.01,
genomeDefinition=netdef,
evaluators=env.fitness,
species=config.defaultSpeciesMethod()
)
runParam = evolution.createRunParam()
evolutionTask = EvolutionTask(1,popParam,env.callback)
evolutionTask.execute(runParam)