Exemplos de DQN_Agent.test em Python

Exemplo n.º 1

Arquivo: run_dp_dqn.py Projeto: zachkeer/Torch-rl

import gym
import time
from agent.DQN import DQN_Agent
from model.Network import DenseNet
from torch import nn
from common.Policy_for_DQN import EpsGreedyQPolicy
#%%
envID = 'D_place_action-v0'
env = gym.make(envID)

nowtime = time.strftime('%y%m%d%H%M', time.localtime())
path = "savedate" + '/' + envID + "-dqn-" + nowtime + '/'
#%%

policy = EpsGreedyQPolicy()
model = DenseNet(env.observation_space.shape[0],
                 env.action_space.n,
                 hidden_activate=nn.Tanh())

Agent = DQN_Agent(env, model, policy, gamma=0.90, lr=1e-3, path=path)

# Agent.train(max_step=1e6, render=False, verbose=2)
# Agent.save_weights(path)
#%%
path = "savedate" + '/' + envID + "-dqn-" + "2002191728"
Agent.load_weights(path)
Agent.test(max_step=10000, render=True, verbose=2)

Exemplo n.º 2

Arquivo: RUN_Catrpole_with_DQN.py Projeto: Seugit/Torch-rl

import gym
import time
from agent.DQN import DQN_Agent
from model.Network import DenseNet
from torch import nn
from common.Policy_for_DQN import BoltzmannQPolicy
#%%
envID = "CartPole-v0"
env = gym.make(envID)
nowtime = time.strftime('%y%m%d%H%M', time.localtime())
path = "savedate" + '/' + envID + "dqn" + nowtime + '/'
#%%

policy = BoltzmannQPolicy()
model = DenseNet(env.observation_space.shape[0],
                 env.action_space.n,
                 hidden_activate=nn.Tanh())

Agent = DQN_Agent(env, model, policy, gamma=0.99, lr=1e-3, path=path)

Agent.train(max_step=100000, render=False, verbose=2)
Agent.test(max_step=10000, render=False, verbose=2)