def save_weight_to_pkl(self):
if not os.path.exists(self.model.weight_dir):
os.makedirs(self.model.weight_dir)
for name in self.w.keys():
save_pkl(self.w[name].eval(),
os.path.join(self.model.weight_dir, "%s.pkl" % name))
def discounted_return(self):
if self.model.solver == 'ValueIteration':
solver = self.solver_factory(self)
self.run_value_iteration(solver, 1)
if self.model.save:
save_pkl(solver.gamma,
os.path.join(self.model.weight_dir,
'VI_planning_horizon_{}.pkl'.format(self.model.planning_horizon)))
elif not self.model.use_tf:
self.multi_epoch()
else:
self.multi_epoch_tf()
print('\n')
console(2, module, 'epochs: ' + str(self.model.n_epochs))
console(2, module, 'ave undiscounted return/step: ' + str(self.experiment_results.undiscounted_return.mean) +
' +- ' + str(self.experiment_results.undiscounted_return.std_err()))
console(2, module, 'ave discounted return/step: ' + str(self.experiment_results.discounted_return.mean) +
' +- ' + str(self.experiment_results.discounted_return.std_err()))
console(2, module, 'ave time/epoch: ' + str(self.experiment_results.time.mean))
self.logger.info('env: ' + self.model.env + '\t' +
'epochs: ' + str(self.model.n_epochs) + '\t' +
'ave undiscounted return: ' + str(self.experiment_results.undiscounted_return.mean) + ' +- ' +
str(self.experiment_results.undiscounted_return.std_err()) + '\t' +
'ave discounted return: ' + str(self.experiment_results.discounted_return.mean) +
' +- ' + str(self.experiment_results.discounted_return.std_err()) +
'\t' + 'ave time/epoch: ' + str(self.experiment_results.time.mean))
def save_alpha_vectors(self):
if not os.path.exists(self.model.weight_dir):
os.makedirs(self.model.weight_dir)
av = self.alpha_vectors()
save_pkl(
av,
os.path.join(self.model.weight_dir,
"linear_alpha_net_vectors.pkl"))