def reset(self):
# core.log("[ai] resetting environment ...")
self._epoch_num = 0
state = self._next_epoch()
self.last['state'] = state
self.last['state_v'] = featurizer.featurize(state, 1)
return self.last['state_v']
def step(self, policy):
# create the parameters from the policy and update
# update them in the algorithm
self._apply_policy(policy)
self._epoch_num += 1
# wait for the algorithm to run with the new parameters
state = self._next_epoch()
self.last['reward'] = state['reward']
self.last['state'] = state
self.last['state_v'] = featurizer.featurize(state, self._epoch_num)
self._agent.on_ai_step()
return self.last['state_v'], self.last['reward'], not self._agent.is_training(), {}