def load_action(path, scope, game, training_flag):
env = game.env
env.set_training_flag(training_flag)
param_path = os.getcwd() + '/network_parameters/param.json'
param = jp.load_json_data(param_path)
act = learn_multi_nets(env,
network=models.mlp(num_layers=param['num_layers'],
num_hidden=param['num_hidden']),
total_timesteps=0,
load_path=path,
scope=scope + '/')
return act
def rand_def_str_generator(env, game):
# Generate random nn for attacker.
num_layers = game.num_layers
num_hidden = game.num_hidden
env.set_training_flag(0)
act_def = learn_multi_nets(env,
network=models.mlp(num_hidden=num_hidden,
num_layers=num_layers - 3),
total_timesteps=0)
print("Saving defender's model to pickle. Epoch in name is equal to 1.")
act_def.save(DIR_def + "def_str_epoch" + str(1) + ".pkl")
# game.def_str.append("def_str_epoch" + str(1) + ".pkl")
game.add_def_str("def_str_epoch" + str(1) + ".pkl")