def __init__(self,
policy,
teacher_env=None,
agent_env=None,
datasource='simulation',
phi=None,
theta=None):
# init
self.datasource = datasource
self.policy = policy
if self.datasource == 'simulation':
self.teacher_env = ffenv.FireflyEnv(
) # init env, need setup later with arg
self.agent_env = ffenv.FireflyEnv()
elif self.datasource == 'behavior':
self.agent_env = ffenv.FireflyEnv()
pass
import matplotlib.pyplot as plt
from Config import Config
from FireflyEnv import ffenv
import numpy as np
from numpy import pi
import matplotlib.pyplot as plt
agent_name = "DDPG_selu_skip_96reward1000000_9 26 16 43"
num_episode = 20
arg = Config()
# arg.gains_range[0:2]=[0.9,0.91]
# arg.std_range=[0.02,0.03,0.02,0.03]
# arg.std_range=[0.0001,0.001,0.0001,0.001]
# arg.gains_range=[0.99,1.,0.99,1.]
env = ffenv.FireflyEnv(arg)
baselines_selu = DDPG.load(agent_name)
torch_model_selu = policy_torch.copy_mlp_weights(baselines_selu,
layers=[256, 256, 64, 32],
act_fn=nn.functional.selu)
torch_model_selu.name = 'selu'
# baselines_relu = DDPG.load("DDPG_theta")
# torch_model_relu = policy_torch.copy_mlp_weights(baselines_relu,layers=[32,64])
# torch_model_relu.name='relu'
agent = torch_model_selu
# create saving vars
all_ep = []
inverse_arg = Inverse_Config()
env_arg = Config()
# policy=DDPG.load("DDPG_ff")
# print(os.getcwd())
phi = reset_theta(inverse_arg.gains_range, inverse_arg.std_range,
inverse_arg.goal_radius_range)
# phi=torch.tensor([1.0537, 0.7328, 0.7053, 1.2038, 0.9661, 0.8689, 0.2930, 1.9330, 0.2000])
theta = reset_theta(inverse_arg.gains_range, inverse_arg.std_range,
inverse_arg.goal_radius_range)
# testing env
teacher_env = ffenv.FireflyEnv(env_arg)
agent_env = ffenv.FireflyEnv(env_arg)
teacher_env.assign_presist_phi(phi)
teacher_env.reset_theta = False
# print(teacher_env.theta)
agent_env.assign_presist_phi(theta)
agent_env.reset_theta = False
# print(agent_env.theta)
# print('state',agent_env.state)
# testing torch agent
import policy_torch
baselines_mlp_model = DDPG.load("DDPG_theta")
policy = policy_torch.copy_mlp_weights(baselines_mlp_model)
from Config import Config
from DDPGv2Agent.rewards import * #reward
arg = Config()
import numpy as np
import time
import torch
from ff_policy.policy_selu import SoftPolicy
import tensorflow as tf
from stable_baselines.common.noise import NormalActionNoise, OrnsteinUhlenbeckActionNoise
action_noise = OrnsteinUhlenbeckActionNoise(mean=np.zeros(2),
sigma=float(0.5) * np.ones(2))
# env=ffenv.FireflyEnv(arg,kwargs={'reward_function':return_reward_location})
env = ffenv.FireflyEnv(arg, kwargs={})
model = DDPG(
MlpPolicy,
env,
verbose=1,
tensorboard_log="./DDPG_tb/",
full_tensorboard_log=False,
action_noise=action_noise,
gamma=0.99,
memory_policy=None,
eval_env=None,
nb_train_steps=50,
nb_rollout_steps=100,
nb_eval_steps=100,
param_noise=None,