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)
agent = TD3Agent(env, save_path=save_path, critic_lr=1e-3,
actor_lr=1e-3).to(device)
# agent=TD3Agent(
# env,
# save_path="./evals/td3_example",
# critic_lr=1e-3,
# actor_lr=1e-3
# ).to(device)
# agent.train(env,100000,train_freq=1,batch_size=100,policy_noise=0.1,policy_noise_clip=.25,gamma=.99, policy_freq=2, tau=0.005,warmup_steps=10000,buffer_size=50000, exp_noise=.1,eval_freq=1, render=True, eval_episodes=1)
agent.train(env,
500,
train_freq=1,
batch_size=800,
policy_noise=0.1,
policy_noise_clip=.25,
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,
# save_path="./evals/td3_example",
# critic_lr=1e-3,
# actor_lr=1e-3
# ).to(device)
# from pyforce.nn import default_network_components
from pyforce.agents import TD3Agent
# import gym
import torch
import sandbox_env
# from pyforce import agents
# device="cuda:0" if torch.cuda.is_available() else "cpu"
# # torch.cuda.set_device(0)
device = "cpu"
# print(device)
env = wrap_openai_gym(sandbox_env.App(always_render=True, verbose=False))
agent = TD3Agent(env,
save_path="./evals/td3_new",
critic_lr=1e-3,
actor_lr=1e-3).to(device)
# agent=TD3Agent(
# env,
# save_path="./evals/td3_example",
# critic_lr=1e-3,
# actor_lr=1e-3
# ).to(device)
# agent.train(env,100000,train_freq=1,batch_size=100,policy_noise=0.1,policy_noise_clip=.25,gamma=.99, policy_freq=2, tau=0.005,warmup_steps=10000,buffer_size=50000, exp_noise=.1,eval_freq=1, render=True, eval_episodes=1)
agent.train(env,
100000,
train_freq=1,
batch_size=800,
policy_noise=0.1,