def eval_genome(genome, config, stage):
global sim, action
global numtimes
global lastcleared
global clearnum
global firstclear
global firstclearnum
global nowstageindex
global nowBest
cleared = set()
net = neat.nn.FeedForwardNetwork.create(genome, config)
#net = neat.nn.RecurrentNetwork.create(genome,config)
trystage = "data/multi" + str(stage) + ".map"
fitness = 0.0
for i in range(1):
sim = snaketrain2.PyAction(0, trystage)
firstxdif = (math.sqrt(abs(sim.map.goalx - sim.map.python.fpx) / 2.0))
checkact = 0
# Run the given simulation for up to num_steps time steps.
bonus = 0.0
hp = 0.0
score = 0.0
clear = False
checkact = 0
prev_action = 0
prev_inputs = []
while sim.game_state != 2 and sim.game_state != 3 and sim.t < simulation_seconds:
if (checkact % frameskip == 0):
inputs = sim.get_displayinfo()
#print(inputs)
prev_inputs = inputs
action = eval_network(net, inputs)
prev_action = action
else:
inputs = prev_inputs
action = prev_action
checkact += 1
# Apply action to the simulated cart-pole
#controll = snaketrain2.discrete_actuator_force(action)
#print(sim.t)
#print(sim.map.python.fpx,sim.map.python.fpy)
#sleep(1)
sim.running(action)
#sleep(1)
#print(sim.game_state)
#if sim.game_state==2 or sim.game_state==3 or sim.t > simulation_seconds:
#break
# Stop if the network fails to keep the cart within the position or angle limits.
# The per-run fitness is the number of time steps the network can balance the pole
# without exceeding these limits.
#print(sim.t)
hp = sim.map.python.HP
score = sim.map.python.score
x = sim.map.python.fpx
y = sim.map.python.fpy
if sim.game_state == 2 or sim.t >= simulation_seconds:
bonus = -10.0
elif sim.game_state == 3:
clear = True
if (clear == True):
global trainend
fitness += 150 + hp + score // 50 + sim.gametime // 60
if (trainend == False):
trainend = True
clearinfo.append((nowtry, 100.0))
with open('Results/multinormalneat' + str(stage) + '.txt',
'a') as f:
f.write("(" + str(nowtry) + "," + str(100.0) + ")")
f.write('\n')
#fitness = sim.map.python.HP/4.0+sim.map.python.score/100.0+bonus-math.sqrt((sim.map.python.rect.x-sim.map.goalx)**2+(sim.map.python.rect.y-sim.map.goaly)**2)/20.0
else:
fitness = fitness + bonus + hp // 20 + score // 50 + firstxdif - (
math.sqrt(abs(sim.map.goalx - x) / 2.0))
#math.sqrt((sim.map.python.rect.x-sim.map.goalx)**2+(sim.map.python.rect.y-sim.map.goaly)**2)/20.0
if (nowBest < (x / sim.map.goalx) * 100):
nowBest = (x / sim.map.goalx) * 100
del sim
#pygame.quit()
#del sim
#print(numtimes,fitness)
#fitnesses.append(fitness)
# The genome's fitness is its worst performance across all runs.
if (len(cleared) > len(lastcleared)):
lastcleared = cleared
return fitness
def eval_genome(genome, config, stage):
global sim, action
global numtimes
global lastcleared
global clearnum
global firstclear
global firstclearnum
global nowstageindex
global nowBest
cleared = set()
net = neat.nn.FeedForwardNetwork.create(genome, config)
#net = neat.nn.RecurrentNetwork.create(genome,config)
trystage = "data/multi" + str(stage) + ".map"
fitness = 0.0
for i in range(1):
sim = snaketrain2.PyAction(0, trystage)
firstxdif = (math.sqrt(abs(sim.map.goalx - sim.map.python.fpx) / 2.0))
checkact = 0
# Run the given simulation for up to num_steps time steps.
bonus = 0.0
hp = 0.0
score = 0.0
clear = False
checkact = 0
prev_action = 0
prev_inputs = []
while sim.game_state != 2 and sim.game_state != 3 and sim.t < simulation_seconds:
if (checkact % 20 == 0):
metainputs = sim.get_metainfo()
choice = eval_network(net, metainputs)
if (choice == 0):
actor = neat.nn.FeedForwardNetwork.create(
NoNets[choice], actorconfig)
elif (choice >= 1 and choice <= 15):
actor = neat.nn.FeedForwardNetwork.create(
EnemyNets[choice - 1], actorconfig)
else:
choice -= 15
actor = neat.nn.FeedForwardNetwork.create(
JumpNets[choice - 1], actorconfig)
if (checkact % frameskip == 0):
inputs = sim.get_displayinfo()
#print(inputs)
prev_inputs = inputs
action = np.argmax(actor.activate(inputs))
prev_action = action
else:
inputs = prev_inputs
action = prev_action
checkact += 1
sim.running(action)
hp = sim.map.python.HP
score = sim.map.python.score
x = sim.map.python.fpx
y = sim.map.python.fpy
if sim.game_state == 2 or sim.t >= simulation_seconds:
bonus = -10.0
elif sim.game_state == 3:
clear = True
if (clear == True):
global trainend
fitness += 150 + hp + score // 50 + sim.gametime // 60
if (trainend == False):
trainend = True
clearinfo.append((nowtry, 100.0))
#with open('Results/multifull'+str(stage)+'.txt','a') as f:
# f.write("("+str(nowtry)+","+str(100.0)+")")
# f.write('\n')
#fitness = sim.map.python.HP/4.0+sim.map.python.score/100.0+bonus-math.sqrt((sim.map.python.rect.x-sim.map.goalx)**2+(sim.map.python.rect.y-sim.map.goaly)**2)/20.0
else:
fitness = fitness + bonus + hp // 20 + score // 50 + firstxdif - (
math.sqrt(abs(sim.map.goalx - x) / 2.0))
#print(hp,score,x,y)
#math.sqrt((sim.map.python.rect.x-sim.map.goalx)**2+(sim.map.python.rect.y-sim.map.goaly)**2)/20.0
if (nowBest < (x / sim.map.goalx) * 100):
nowBest = (x / sim.map.goalx) * 100
del sim
#pygame.quit()
#del sim
#print(numtimes,fitness)
#fitnesses.append(fitness)
# The genome's fitness is its worst performance across all runs.
if (len(cleared) > len(lastcleared)):
lastcleared = cleared
return fitness