import torch
import torch.nn as nn
import torch.nn.functional as F
import math
import sys
import random
#create seed
from utils.tool import read_seed
torch.manual_seed(read_seed())
class InitLinear(nn.Module):
def __init__(self, in_features, out_features):
super(InitLinear, self).__init__()
self.in_features = in_features
self.out_features = out_features
self.weight = nn.Parameter(torch.empty(out_features, in_features))
self.bias = nn.Parameter(torch.empty(out_features))
self.reset_parameters()
def reset_parameters(self):
mu_range = 1.0 / math.sqrt(self.in_features)
self.weight.data.uniform_(-mu_range, mu_range)
self.bias.data.uniform_(-mu_range, mu_range)
def forward(self, inp):
return F.linear(inp, self.weight, self.bias)
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
from collections import deque
import random
import time
import sys
from Models.Networks.SimpleNetwork import Network
from utils.tool import read_config, read_seed
np.random.seed(read_seed())
config = read_config("./Configs/configD2QN.yml")
N_ACTIONS = config["n_actions"]
GAMMA = config["gamma"]
EPSILON_DECAY = config["epsilon_decay"]
EPSILON_MIN = config["epsilon_min"]
BATCH_SIZE = config["batch_size"]
MIN_REPLAY_SIZE = config["min_replay_size"]
TARGET_UPDATE = config["target_update"]
SAMPLING_TIME = config["sampling_time"]
NAME = config["model_name"]
# Deep Q Network agent
class Agent:
def __init__(self):
parser = argparse.ArgumentParser(
description='choose your model')
parser.add_argument('-model', type=str, default='DQN', help='DQN/D2QN/D3QN/Noisy/PER/ApeX')
parser.add_argument('-seed', type=int, default=-1, help='save/init')
args = parser.parse_args()
#seeds
if args.seed == -1:
config = read_config("./Configs/config" + args.model + ".yml")
write_seed(config["seed"])
else:
write_seed(args.seed)
print("seed:", read_seed())
#Models
if args.model == "ApeX":
runApeX()
elif args.model == "DQN":
env = EnvCarla.CarEnv(0)
agent = ModelDQN.Agent()
TrainModel.run(config, agent, env)
elif args.model == "D2QN":
env = EnvCarla.CarEnv(0)
agent = ModelD2QN.Agent()
TrainModel.run(config, agent, env)
elif args.model == "D3QN":
