def act(self, stateObs, _):
global model
actionID, _ = model.predict(transformFrame(stateObs, x=110, y=110))
if actionID >= 4:
actionID = 0
return actionID
def reset(self):
# This doesn't work if each environment has a different size
for remote in self.vecenv.remotes:
remote.send(('reset', None))
resetFrames = [remote.recv() for remote in self.vecenv.remotes]
returnList = []
for frame in resetFrames:
returnList.append(transformFrame(frame, x=self.x, y=self.y))
return np.stack(returnList)
def act(self, stateObs, actions):
print(actions)
global model
actionID, _ = model.predict(transformFrame(stateObs, x=110, y=110))
if actionID >= len(actions):
actionID = 0
return actionID
def step_wait(self):
obsList, rewsList, donesList, infosList = ([] for i in range(4))
for remote in self.vecenv.remotes:
(ob, rew, done, info) = remote.recv()
remote.send(('env_method', ('__str__', {}, {})))
string = remote.recv()
if not rew == 0: # Don't need to waste time normalizing 0s
rew = normalizeReward(rew, string)
rewsList.append(rew)
obsList.append(transformFrame(ob, x=self.x, y=self.y))
donesList.append(done)
infosList.append(info)
self.vecenv.waiting = False
return np.stack(obsList), np.stack(rewsList), np.stack(donesList), np.stack(infosList)
def act(self, stateObs, _):
global model
actionID, _ = model.predict(transformFrame(stateObs, x=300, y=300))
return actionID