file_path = agent_params['save_path'] + '/params'
with open(file_path, 'wb') as f:
pickle.dump(parm_list, f)
file_path_txt = file_path + '.txt'
with open(file_path_txt, 'w') as f:
f.write('agent_params\n')
for k in agent_params.keys():
f.write('\t' + k + ' = ' + str(agent_params[k]) + '\n')
f.write('env_params\n')
for k in sandbox_env.env_params.keys():
f.write('\t' + k + ' = ' + str(sandbox_env.env_params[k]) + '\n')
# device="cuda:0" if torch.cuda.is_available() else "cpu"
# torch.cuda.set_device(0)
device= "cpu"
env=wrap_openai_gym(sandbox_env.App(always_render=False))
agent=TD3Agent(
env,
save_path=agent_params['save_path'],
critic_lr=agent_params['critic_lr'],
actor_lr=agent_params['actor_lr']
).to(device)
if LOAD_AGENT_FROM is not None:
with open(LOAD_AGENT_FROM, 'rb') as f:
state_dict = pickle.load(f)
agent.load_state_dict(state_dict)
# agent=TD3Agent(
# env,
from pyforce.env import wrap_openai_gym
from pyforce.nn import default_network_components
from pyforce.agents import A2CAgent
import gym
import torch
device = "cuda:0" if torch.cuda.is_available() else "cpu"
env = wrap_openai_gym(gym.make("LunarLanderContinuous-v2"))
observation_processor, hidden_layers, action_mapper = default_network_components(
env)
agent = A2CAgent(observation_processor,
hidden_layers,
action_mapper,
save_path="./evals/a2c_example",
value_lr=1e-3,
policy_lr=1e-3).to(device)
agent.train(env,
episodes=1000,
train_freq=256,
eval_freq=50,
render=True,
gamma=.99,
entropy_coef=.01)
with open(file_path, 'wb') as f:
pickle.dump(parm_list, f)
file_path_txt = file_path + '.txt'
with open(file_path_txt, 'w') as f:
f.write('agent_params\n')
for k in agent_params.keys():
f.write('\t' + k + ' = ' + str(agent_params[k]) + '\n')
f.write('env_params\n')
for k in gauss_env.env_params.keys():
f.write('\t' + k + ' = ' + str(gauss_env.env_params[k]) + '\n')
# device="cuda:0" if torch.cuda.is_available() else "cpu"
# # torch.cuda.set_device(0)
device = "cpu"
env = wrap_openai_gym(gauss_env.App(always_render=False, verbose=False))
observation_processor, hidden_layers, action_mapper = default_network_components(
env)
agent = PPOAgent(observation_processor,
hidden_layers,
action_mapper,
save_path=agent_params['save_path'],
value_lr=agent_params['value_lr'],
policy_lr=agent_params['policy_lr']).to(device)
if LOAD_AGENT_FROM is not None:
with open(LOAD_AGENT_FROM, 'rb') as f:
state_dict = pickle.load(f)
agent.load_state_dict(state_dict)
LOAD_PATH = './evals/ppo_gauss/4/'
file_path = LOAD_PATH + '/params'
with open(file_path, 'rb') as f:
parm_list = pickle.load(f)
agent_params = parm_list[0]
env_parms = parm_list[1]
gauss_env.N_BOTS = env_parms['N_BOTS']
# device="cuda:0" if torch.cuda.is_available() else "cpu"
# # torch.cuda.set_device(0)
device = "cpu"
env=wrap_openai_gym(gauss_env.App(always_render=True, plot_reward=False, verbose=False))
observation_processor,hidden_layers,action_mapper=default_network_components(env)
agent=PPOAgent(
observation_processor,
hidden_layers,
action_mapper,
save_path=agent_params['save_path'],
value_lr=agent_params['value_lr'],
policy_lr=agent_params['policy_lr']
).to(device)
file_path = LOAD_PATH + '/agent'
# file_path = LOAD_PATH + '/agent4903861.0'
with open(file_path, 'rb') as f:
# LOAD_PATH = './evals/preliminary_minimal/reward_function/distance-True/2/'
file_path = LOAD_PATH + '/params'
with open(file_path, 'rb') as f:
parm_list = pickle.load(f)
agent_params = parm_list[0]
env_parms = parm_list[1]
sandbox_env.N_BOTS = env_parms['N_BOTS']
# device="cuda:0" if torch.cuda.is_available() else "cpu"
# # torch.cuda.set_device(0)
device = "cpu"
env = wrap_openai_gym(sandbox_env.App(traj_savepath=LOAD_PATH))
if PPO:
observation_processor, hidden_layers, action_mapper = default_network_components(
env)
agent = PPOAgent(observation_processor,
hidden_layers,
action_mapper,
save_path=agent_params['save_path'],
value_lr=agent_params['value_lr'],
policy_lr=agent_params['policy_lr']).to(device)
else:
agent = TD3Agent(env,
import torch
from pathlib import Path
import minimal_v4 as envi
# import minimal_v1
# from pyforce import agents
save_path = "./evals/td3_example"
description = 'Minimal 4: Umgebung im Bereich von (-100,-100) bis (100,100), Observationspace: Eigene Koordinaten und Zielposition, Startposition(0,0), Zufällige Zielposition'
# description = 'minimal_v4'
device = "cuda:0" if torch.cuda.is_available() else "cpu"
# # torch.cuda.set_device(0)
# device= "cpu"
# print(device)
# env=wrap_openai_gym(minimal_v1.App())
env = wrap_openai_gym(envi.App())
Path(save_path).mkdir(exist_ok=True)
run_number = 0
for p in Path(save_path).iterdir():
if p.is_dir() and p.name.isnumeric():
if int(p.name) > run_number:
run_number = int(str(p.name))
run_number += 1
save_path += ('/' + str(run_number))
Path(save_path).mkdir(exist_ok=True)
with open(save_path + '/description.txt', mode='w') as f:
f.write(description)
LOAD_PATH = './evals/ppo_pycking/4/'
file_path = LOAD_PATH + '/params'
with open(file_path, 'rb') as f:
parm_list = pickle.load(f)
agent_params = parm_list[0]
env_parms = parm_list[1]
pycking_env3.N_BOTS = env_parms['N_BOTS']
# device="cuda:0" if torch.cuda.is_available() else "cpu"
# # torch.cuda.set_device(0)
device = "cpu"
env=wrap_openai_gym(pycking_env3.App(always_render=True, verbose=False))
observation_processor,hidden_layers,action_mapper=default_network_components(env)
agent=PPOAgent(
observation_processor,
hidden_layers,
action_mapper,
save_path=agent_params['save_path'],
value_lr=agent_params['value_lr'],
policy_lr=agent_params['policy_lr']
).to(device)
file_path = LOAD_PATH + '/agent'
with open(file_path, 'rb') as f:
state_dict = pickle.load(f)
LOAD_PATH = './evals/ppo_sandbox/21'
file_path = LOAD_PATH + '/params'
with open(file_path, 'rb') as f:
parm_list = pickle.load(f)
agent_params = parm_list[0]
env_parms = parm_list[1]
sandbox_env.N_BOTS = env_parms['N_BOTS']
# device="cuda:0" if torch.cuda.is_available() else "cpu"
# # torch.cuda.set_device(0)
device = "cpu"
env = wrap_openai_gym(sandbox_env.App(always_render=True, verbose=False))
if PPO:
observation_processor, hidden_layers, action_mapper = default_network_components(
env)
agent = PPOAgent(observation_processor,
hidden_layers,
action_mapper,
save_path=agent_params['save_path'],
value_lr=agent_params['value_lr'],
policy_lr=agent_params['policy_lr']).to(device)
else:
agent = TD3Agent(env,
file_path = agent_params['save_path'] + '/params'
with open(file_path, 'wb') as f:
pickle.dump(parm_list, f)
file_path_txt = file_path + '.txt'
with open(file_path_txt, 'w') as f:
f.write('agent_params\n')
for k in agent_params.keys():
f.write('\t' + k + ' = ' + str(agent_params[k]) + '\n')
f.write('env_params\n')
for k in preliminary_env.env_params.keys():
f.write('\t' + k + ' = ' + str(preliminary_env.env_params[k]) +
'\n')
device = "cpu"
env = wrap_openai_gym(
preliminary_env.App(always_render=False, verbose=False))
# env=wrap_openai_gym(minimal_preliminary.App())
if ag == 'ppo':
observation_processor, hidden_layers, action_mapper = default_network_components(
env)
agent = PPOAgent(observation_processor,
hidden_layers,
action_mapper,
save_path=agent_params['save_path'],
value_lr=agent_params['value_lr'],
policy_lr=agent_params['policy_lr']).to(device)
agent.train(env,
episodes=agent_params['episodes'],
train_freq=agent_params['train_freq'],
LOAD_PATH = './evals/final/21/'
# LOAD_PATH = './evals/algorithm/ppo/3/'
file_path = LOAD_PATH + '/params'
with open(file_path, 'rb') as f:
parm_list = pickle.load(f)
agent_params = parm_list[0]
env_parms = parm_list[1]
# sandbox_env.N_BOTS = env_parms['N_BOTS']
# device="cuda:0" if torch.cuda.is_available() else "cpu"
# # torch.cuda.set_device(0)
device = "cpu"
env=wrap_openai_gym(sandbox_env.App(always_render=True, verbose=False, traj_savepath=LOAD_PATH))
if PPO:
observation_processor,hidden_layers,action_mapper=default_network_components(env)
agent=PPOAgent(
observation_processor,
hidden_layers,
action_mapper,
save_path=agent_params['save_path'],
value_lr=agent_params['value_lr'],
policy_lr=agent_params['policy_lr']
).to(device)
else: