d59156c9157ff3aab5b3cb24b3cc65bb9269004c,ml/rl/test/gridworld/test_gridworld_sac.py,TestGridworldSAC,_test_sac_trainer,#TestGridworldSAC#Any#Any#,125

Before Change

    def _test_sac_trainer(self, use_2_q_functions=False, use_gpu=False):
        environment = GridworldContinuous()
        samples = environment.generate_samples(100000, 0.25, DISCOUNT)
        trainer = self.get_sac_trainer(
            environment, self.get_sac_parameters(use_2_q_functions), use_gpu
        )
        // evaluator = GridworldSACEvaluator(environment, True, DISCOUNT, False, samples)
        evaluator = GridworldContinuousEvaluator(
            environment,
            assume_optimal_policy=False,
            gamma=DISCOUNT,
            use_int_features=False,
        )

        // FIXME: need to be able to export w/o calling .cpu()
        // critic_predictor = self.get_predictor(trainer, environment)
        // self.assertGreater(evaluator.evaluate(critic_predictor), 0.15)

        tdps = environment.preprocess_samples(
            samples, self.minibatch_size, use_gpu=use_gpu
        )

        // critic_predictor = trainer.predictor(actor=False)

        // evaluator.evaluate_critic(critic_predictor)

After Change

        self.test_save_load = False
        self.tolerance_threshold = 0.12
        if use_gpu:
            self.run_pre_training_eval = False
        self.evaluate_gridworld(environment, evaluator, trainer, exporter, use_gpu)

        // Make sure actor predictor works
        actor_predictor = self.get_actor_predictor(trainer, environment)

Instances