def __init__(self,
index,
model,
env,
memory,
lr=1e-2,
preprocess=T.ToTensor(),
path=None,
frompath=None,
num_episodes=1000,
epsend=0.05,
epsstart=0.9,
epsdecay=200,
k=4,
strategy='future'):
self.index = index
self.model = model
self.envstr = env
self.env = gym.make(self.envstr)
self.env.reset()
self.memory = memory
self.lr = lr
#self.optimizer = optim.RMSprop(self.model.parameters(), lr=self.lr )
self.optimizer = optim.Adam(self.model.parameters(),
lr=self.lr,
eps=1.5e-4)
bashlogger.info('Optimizer {}: ok.'.format(self.index))
self.preprocess = preprocess
self.path = path
self.frompath = frompath
self.num_episodes = num_episodes
self.epsend = epsend
self.epsstart = epsstart
self.epsdecay = epsdecay
#HER params :
self.k = k
self.strategy = strategy
self.sl = statsLogger(path=self.path,
filename='logs{}.csv'.format(self.index))
self.workerfn = lambda: train(model=self.model,
env=self.env,
memory=self.memory,
optimizer=self.optimizer,
logger=self.sl,
preprocess=self.preprocess,
path=self.path,
frompath=self.frompath,
num_episodes=self.num_episodes,
epsend=self.epsend,
epsstart=self.epsstart,
epsdecay=self.epsdecay,
k=self.k,
strategy=self.strategy)
self.thread = threading.Thread(target=self.workerfn)
def __init__(self,index,model,env,memory,lr=1e-2,preprocess=T.ToTensor(),path=None,frompath=None,num_episodes=1000,epsend=0.05,epsstart=0.9,epsdecay=200,TAU=1e-3,k=4,strategy='future') :
self.index = index
self.model = model
self.wmodel = copy.deepcopy(model)
hard_update(self.wmodel,self.model)
global use_cuda
if use_cuda :
self.wmodel = self.wmodel.cuda()
self.envstr = env
self.env = gym.make(self.envstr)
self.env.reset()
self.memory = memory
self.lr = lr
self.TAU = TAU
self.optimizer = optim.Adam(self.model.parameters(), lr=self.lr )
bashlogger.info('Optimizer {}: ok.'.format(self.index) )
self.preprocess = preprocess
self.path = path
self.frompath = frompath
self.num_episodes = num_episodes
self.epsend = epsend
self.epsstart = epsstart
self.epsdecay = epsdecay
#HER params :
self.k = k
self.strategy = strategy
self.sl = statsLogger(path=self.path,filename='logs{}.csv'.format(self.index) )
self.workerfn = lambda: self.train(model=self.wmodel,
env=self.env,
memory=self.memory,
optimizer=self.optimizer,
logger=self.sl,
preprocess=self.preprocess,
path=self.path,
frompath=self.frompath,
num_episodes=self.num_episodes,
epsend=self.epsend,
epsstart=self.epsstart,
epsdecay=self.epsdecay,
k=self.k,
strategy=self.strategy,
singlegoal=False)
self.thread = threading.Thread(target=self.workerfn)
def __init__(self,index,model,env,memory,preprocess=T.ToTensor(),path=None,frompath=None,num_episodes=1000,nbr_ep_per_train=1,HER={'use_her':True,'k':4,'strategy':'future','singlegoal':False},use_cuda=True,rendering=False) :
self.index = index
self.model = model
self.envstr = env
self.env = gym.make(self.envstr)
self.env.reset()
self.memory = memory
self.optimizers = self.model.generate_optimizers()
self.preprocess = preprocess
self.path = path
self.frompath = frompath
self.num_episodes = num_episodes
self.nbr_ep_per_train = nbr_ep_per_train
#HER params :
self.HER = HER
self.use_cuda = use_cuda
self.rendering = rendering
self.sl = statsLogger(path=self.path,filename='logs{}.csv'.format(self.index) )
#self.workerfn = lambda: self.trainIN( index=self.index,
self.workerfn = lambda: self.trainOUT( index=self.index,
model=self.model,
env=self.env,
memory=self.memory,
optimizers=self.optimizers,
logger=self.sl,
preprocess=self.preprocess,
path=self.path,
frompath=self.frompath,
num_episodes=self.num_episodes,
HER=self.HER,
use_cuda=self.use_cuda,
rendering=self.rendering)
self.thread = threading.Thread(target=self.workerfn)