def __init__(self, config):
with open(config, 'r') as f:
config = json.load(f)
self.episode_length = config['train']['episode_length']
self.env = gym.make(config['model']['gym'])
state_size = config['model']['state_size']
action_size = config['model']['action_size']
self.policy_net = Policy1D(state_size, action_size)
self.value_net = Value1D(state_size)
self.plosses = []
self.vlosses = []
self.stand_time = []
if torch.cuda.is_available():
self.policy_net.cuda()
self.value_net.cuda()
self.device = torch.device("cuda")
print("Using GPU")
else:
self.device = torch.device("cpu")
print("No GPU detected")
self.write_interval = config['model']['write_interval']
self.train_info_path = config['model']['trainer_save_path']
self.policy_path = config['model']['policy_save_path'].split('.pt')[0]
self.value_path = config['model']['value_save_path'].split('.pt')[0]
def __init__(self, config):
with open(config, 'r') as f:
config = json.load(f)
self.env = gym.make(config['model']['gym'])
self.mtcar = (config['model']['gym'] == 'MountainCar-v0')
state_size = config['model']['state_size']
action_size = config['model']['action_size']
hidden_size = config['model']['hidden_size']
layer_size = config['model']['hidden_layers']
logheat = config['model']['logheat']
self.policy_net = Policy1D(state_size,
action_size,
hidden_size=hidden_size,
layers=layer_size,
logheat=logheat)
if torch.cuda.is_available():
self.policy_net.cuda()
self.value_net.cuda()
self.device = torch.device("cuda")
print("Using GPU")
else:
self.device = torch.device("cpu")
print("No GPU detected")
self.visual = True
self.cut = config['train']['cutearly']
self.write_interval = config['model']['write_interval']
self.train_info_path = config['model']['trainer_save_path']
self.policy_path = config['model']['policy_save_path'].split('.pt')[0]
def __init__(self, config):
with open(config, 'r') as f:
config = json.load(f)
self.env = gym.make(config['model']['gym'])
state_size = config['model']['state_size']
action_size = config['model']['action_size']
hidden_size = config['model']['hidden_size']
layer_size = config['model']['hidden_layers']
self.action_size = action_size
self.policy_net = Policy1D(state_size,
action_size,
hidden_size=hidden_size,
layers=layer_size)
self.value_net = Value1D(state_size,
hidden_size=hidden_size,
layers=layer_size)
if torch.cuda.is_available():
self.policy_net.cuda()
self.value_net.cuda()
self.device = torch.device("cuda")
print("Using GPU")
else:
self.device = torch.device("cpu")
print("No GPU detected")
self.write_interval = config['model']['write_interval']
self.train_info_path = config['model']['trainer_save_path']
self.policy_path = config['model']['policy_save_path'].split('.pt')[0]
self.value_path = config['model']['value_save_path'].split('.pt')[0]
self.gif_path = config['model']['gif_save_path'].split('.gif')[0]
self.graph_path = config['model']['graph_save_path']
def __init__(self, config):
with open(config, 'r') as f:
config = json.load(f)
self.epochs = config['train']['epochs']
self.env_samples = config['train']['env_samples']
self.episode_length = config['train']['episode_length']
self.gamma = config['train']['gamma']
self.value_epochs = config['train']['value_epochs']
self.policy_epochs = config['train']['policy_epochs']
self.batch_size = config['train']['batch_size']
self.policy_batch_size = config['train']['policy_batch_size']
epsilon = config['train']['epsilon']
# self.env = gym.make(config['model']['gym'])
env_factory = CartPoleEnvironmentFactory()
self.env_threads=1
self.envs = [env_factory.new() for _ in range(self.env_threads)]
rollouts_per_thread = self.env_samples // self.env_threads
remainder = self.env_samples % self.env_threads
self.rollout_nums = ([rollouts_per_thread + 1] * remainder) + ([rollouts_per_thread] * (self.env_threads - remainder))
state_size = config['model']['state_size']
action_size = config['model']['action_size']
self.action_size = action_size
self.policy_net = Policy1D(state_size, action_size)
# self.value_net = Value1D(state_size)
self.value_loss = nn.MSELoss()
self.ppoloss = PPOLoss(epsilon)
self.ppo_low_bnd = 1 - epsilon
self.ppo_up_bnd = 1 + epsilon
betas = (config['train']['betas1'], config['train']['betas2'])
weight_decay = config['train']['weight_decay']
lr = config['train']['lr']
# params = chain(self.policy_net.parameters(), self.value_net.parameters())
self.optim = optim.Adam(self.policy_net.parameters(), lr=lr, betas=betas, weight_decay=weight_decay)
self.plosses = []
self.vlosses = []
self.stand_time = []
if torch.cuda.is_available():
self.policy_net.cuda()
self.value_net.cuda()
self.device = torch.device("cuda")
print("Using GPU")
else:
self.device = torch.device("cpu")
print("No GPU detected")
self.write_interval = config['model']['write_interval']
self.train_info_path = config['model']['trainer_save_path']
self.policy_path = config['model']['policy_save_path'].split('.pt')[0]
self.value_path = config['model']['value_save_path'].split('.pt')[0]
def __init__(self, config):
with open(config, 'r') as f:
config = json.load(f)
self.epochs = config['train']['epochs']
self.env_samples = config['train']['env_samples']
self.episode_length = config['train']['episode_length']
self.gamma = config['train']['gamma']
self.value_epochs = config['train']['value_epochs']
self.policy_epochs = config['train']['policy_epochs']
self.batch_size = config['train']['batch_size']
self.policy_batch_size = config['train']['policy_batch_size']
epsilon = config['train']['epsilon']
self.env = gym.make(config['model']['gym'])
state_size = config['model']['state_size']
action_size = config['model']['action_size']
self.action_size = action_size
self.policy_net = Policy1D(state_size, action_size)
self.value_net = Value1D(state_size)
self.value_loss = nn.MSELoss()
self.ppoloss = PPOLoss(epsilon)
policy_lr = config['train']['policy_lr']
value_lr = config['train']['value_lr']
policy_decay = config['train']['policy_decay']
value_decay = config['train']['value_decay']
self.policy_optim = optim.Adam(self.policy_net.parameters(), lr=policy_lr, weight_decay=policy_decay)
self.value_optim = optim.Adam(self.value_net.parameters(), lr=value_lr, weight_decay=value_decay)
self.plosses = []
self.vlosses = []
self.stand_time = []
if torch.cuda.is_available():
self.policy_net.cuda()
self.value_net.cuda()
self.device = torch.device("cuda")
print("Using GPU")
else:
self.device = torch.device("cpu")
print("No GPU detected")
self.write_interval = config['model']['write_interval']
self.train_info_path = config['model']['trainer_save_path']
self.policy_path = config['model']['policy_save_path'].split('.pt')[0]
self.value_path = config['model']['value_save_path'].split('.pt')[0]
def __init__(self, config):
with open(config, 'r') as f:
config = json.load(f)
self.epochs = config['train']['epochs']
self.value_epochs = config['train']['value_epochs']
self.policy_epochs = config['train']['policy_epochs']
self.policy_batch_size = config['train']['policy_batch_size']
state_size = config['model']['state_size']
action_size = config['model']['action_size']
self.action_size = action_size
self.policy_net = Policy1D(state_size, action_size)
self.value_loss = nn.MSELoss()
epsilon = config['train']['epsilon']
self.ppoloss = PPOLoss(epsilon)
self.ppo_low_bnd = 1 - epsilon
self.ppo_up_bnd = 1 + epsilon
betas = (config['train']['betas1'], config['train']['betas2'])
weight_decay = config['train']['weight_decay']
lr = config['train']['lr']
# params = chain(self.policy_net.parameters(), self.value_net.parameters())
self.optim = optim.Adam(self.policy_net.parameters(),
lr=lr,
betas=betas,
weight_decay=weight_decay)
self.plosses = []
self.vlosses = []
self.avg_rewards = []
self.stand_time = []
if torch.cuda.is_available():
self.policy_net.cuda()
self.value_net.cuda()
self.device = torch.device("cuda")
print("Using GPU")
else:
self.device = torch.device("cpu")
print("No GPU detected")
env = gym.make(config['model']['gym'])
env_samples = config['train']['env_samples']
episode_length = config['train']['episode_length']
gamma = config['train']['gamma']
self.rollFact = RolloutFactory(env,
config['model']['gym'],
self.policy_net,
env_samples,
episode_length,
gamma,
cutearly=config['train']['cutearly'])
self.write_interval = config['model']['write_interval']
self.train_info_path = config['model']['trainer_save_path']
self.policy_path = config['model']['policy_save_path'].split('.pt')[0]
self.value_path = config['model']['value_save_path'].split('.pt')[0]
self.graph_path = config['model']['graph_save_path'].split('.png')[0]