MTRF/algorithms/softlearning/algorithms/multi_sac.py [1039:1078]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
            training_metrics_per_policy = self._evaluate_rollouts(
                training_paths_per_policy, training_environment)
            gt.stamp('training_metrics')

            if evaluation_paths_per_policy:
                evaluation_metrics_per_policy = self._evaluate_rollouts(
                    evaluation_paths_per_policy, evaluation_environment)
                gt.stamp('evaluation_metrics')
            else:
                evaluation_metrics_per_policy = [{} for _ in range(self._num_goals)]

            self._epoch_after_hook(training_paths_per_policy)
            gt.stamp('epoch_after_hook')

            t0 = time.time()

            sampler_diagnostics_per_policy = [
                self._samplers[i].get_diagnostics() for i in range(self._num_goals)]

            diagnostics = self.get_diagnostics(
                iteration=self._total_timestep,
                batches=self._evaluation_batches(),
                training_paths_per_policy=training_paths_per_policy,
                evaluation_paths_per_policy=evaluation_paths_per_policy)

            time_diagnostics = gt.get_times().stamps.itrs

            print("Basic diagnostics: ", time.time() - t0)
            print("Sample count: ", self._sample_count)

            diagnostics.update(OrderedDict((
                *(
                    (f'times/{key}', time_diagnostics[key][-1])
                    for key in sorted(time_diagnostics.keys())
                ),
                ('epoch', self._epoch),
                ('timestep', self._timestep),
                ('timesteps_total', self._total_timestep),
                ('train-steps', self._num_train_steps),
            )))
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



MTRF/algorithms/softlearning/algorithms/phased_sac.py [787:826]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
            training_metrics_per_policy = self._evaluate_rollouts(
                training_paths_per_policy, training_environment)
            gt.stamp('training_metrics')

            if evaluation_paths_per_policy:
                evaluation_metrics_per_policy = self._evaluate_rollouts(
                    evaluation_paths_per_policy, evaluation_environment)
                gt.stamp('evaluation_metrics')
            else:
                evaluation_metrics_per_policy = [{} for _ in range(self._num_goals)]

            self._epoch_after_hook(training_paths_per_policy)
            gt.stamp('epoch_after_hook')

            t0 = time.time()

            sampler_diagnostics_per_policy = [
                self._samplers[i].get_diagnostics() for i in range(self._num_goals)]

            diagnostics = self.get_diagnostics(
                iteration=self._total_timestep,
                batches=self._evaluation_batches(),
                training_paths_per_policy=training_paths_per_policy,
                evaluation_paths_per_policy=evaluation_paths_per_policy)

            time_diagnostics = gt.get_times().stamps.itrs

            print("Basic diagnostics: ", time.time() - t0)
            print("Sample count: ", self._sample_count)

            diagnostics.update(OrderedDict((
                *(
                    (f'times/{key}', time_diagnostics[key][-1])
                    for key in sorted(time_diagnostics.keys())
                ),
                ('epoch', self._epoch),
                ('timestep', self._timestep),
                ('timesteps_total', self._total_timestep),
                ('train-steps', self._num_train_steps),
            )))
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -