def run():
# Main loop
#values_hist = [np.ravel(values.copy())]
actions_executed = 0
last_action_time = 0
position = env.getState().copy()
in_end_position = False
while actions_executed < NUM_ITERATIONS:
if not in_end_position:
# stimulate new state
nest.SetStatus(nest.GetConnections(stimulus, states[position['x']][position['y']]), {'weight': 0.})
position = env.getState().copy()
nest.SetStatus(nest.GetConnections(stimulus, states[position['x']][position['y']]), {'weight': 1.})
nest.SetStatus(wta_noise, {'rate': 3000.})
for t in range(8):
nest.Simulate(5)
time.sleep(0.01)
max_rate = -1
chosen_action = -1
for i in range(len(sd_actions)):
rate = len([e for e in nest.GetStatus([sd_actions[i]], keys='events')[0]['times'] if e > last_action_time]) # calc the "firerate" of each actor population
if rate > max_rate:
max_rate = rate # the population with the hightes rate wins
chosen_action = i
nest.SetStatus(stimulus, {'rate': 5000.})
possible_actions = env.get_possible_actions()
new_position, outcome, in_end_position = env.move(possible_actions[chosen_action])
nest.SetStatus(wta_noise, {'rate': 0.})
for t in range(4):
nest.Simulate(5)
time.sleep(0.01)
last_action_time += 60
actions_executed += 1
else:
position = env.get_agent_pos().copy()
_, in_end_position = env.init_new_trial()
nest.SetStatus(nest.GetConnections(stimulus, states[position['x']][position['y']]), {'weight': 0.})
import time
import os
import json
import mpi_environment as env
policy_filename = "pong_policy.dat"
values_filename = "pong_values.dat"
alpha = 0.1 # values / critic learning parameter
beta = 0.1 # actor learning parameter
gamma = 0.9 # error signal: future states parameter
world_dim = env.getWorldDim()
world_dim = {'y': world_dim[0], 'x': world_dim[1]}
num_possible_moves = env.getActionDim()
state = env.getState()
pol_file = None
val_file = None
if os.path.exists(policy_filename):
pol_file = open(policy_filename, 'r+')
policy = numpy.array(json.loads(pol_file.read()))
pol_file.close()
else:
#create random policy
print num_possible_moves
policy = numpy.random.rand(world_dim['y'], world_dim['x'], num_possible_moves)
#pol_file = open(policy_filename, 'w+')
