Python tune.Trainable方法代码示例

# 需要导入模块: from ray import tune [as 别名]
# 或者: from ray.tune import Trainable [as 别名]
def testAutoregisterTrainable(self):
        def train(config, reporter):
            for i in range(100):
                reporter(timesteps_total=i)

        class B(Trainable):
            def step(self):
                return {"timesteps_this_iter": 1, "done": True}

        register_trainable("f1", train)
        trials = run_experiments({
            "foo": {
                "run": train,
            },
            "bar": {
                "run": B
            }
        })
        for trial in trials:
            self.assertEqual(trial.status, Trial.TERMINATED) 

# 需要导入模块: from ray import tune [as 别名]
# 或者: from ray.tune import Trainable [as 别名]
def testCheckpointAtEnd(self):
        class train(Trainable):
            def step(self):
                return {"timesteps_this_iter": 1, "done": True}

            def save_checkpoint(self, path):
                checkpoint = os.path.join(path, "checkpoint")
                with open(checkpoint, "w") as f:
                    f.write("OK")
                return checkpoint

        trials = run_experiments({
            "foo": {
                "run": train,
                "checkpoint_at_end": True
            }
        })
        for trial in trials:
            self.assertEqual(trial.status, Trial.TERMINATED)
            self.assertTrue(trial.has_checkpoint()) 

# 需要导入模块: from ray import tune [as 别名]
# 或者: from ray.tune import Trainable [as 别名]
def testExportFormats(self):
        class train(Trainable):
            def step(self):
                return {"timesteps_this_iter": 1, "done": True}

            def _export_model(self, export_formats, export_dir):
                path = os.path.join(export_dir, "exported")
                with open(path, "w") as f:
                    f.write("OK")
                return {export_formats[0]: path}

        trials = run_experiments({
            "foo": {
                "run": train,
                "export_formats": ["format"]
            }
        })
        for trial in trials:
            self.assertEqual(trial.status, Trial.TERMINATED)
            self.assertTrue(
                os.path.exists(os.path.join(trial.logdir, "exported"))) 

# 需要导入模块: from ray import tune [as 别名]
# 或者: from ray.tune import Trainable [as 别名]
def testInvalidExportFormats(self):
        class train(Trainable):
            def step(self):
                return {"timesteps_this_iter": 1, "done": True}

            def _export_model(self, export_formats, export_dir):
                ExportFormat.validate(export_formats)
                return {}

        def fail_trial():
            run_experiments({
                "foo": {
                    "run": train,
                    "export_formats": ["format"]
                }
            })

        self.assertRaises(TuneError, fail_trial) 

# 需要导入模块: from ray import tune [as 别名]
# 或者: from ray.tune import Trainable [as 别名]
def testCustomResources(self):
        ray.shutdown()
        ray.init(resources={"hi": 3})

        class train(Trainable):
            def step(self):
                return {"timesteps_this_iter": 1, "done": True}

        trials = run_experiments({
            "foo": {
                "run": train,
                "resources_per_trial": {
                    "cpu": 1,
                    "custom_resources": {
                        "hi": 2
                    }
                }
            }
        })
        for trial in trials:
            self.assertEqual(trial.status, Trial.TERMINATED) 

# 需要导入模块: from ray import tune [as 别名]
# 或者: from ray.tune import Trainable [as 别名]
def testResetTrial(self):
        """Tests that reset works as expected."""

        class B(Trainable):
            def step(self):
                return dict(timesteps_this_iter=1, done=True)

            def reset_config(self, config):
                self.config = config
                return True

        trials = self.generate_trials({
            "run": B,
            "config": {
                "foo": 0
            },
        }, "grid_search")
        trial = trials[0]
        self.trial_executor.start_trial(trial)
        exists = self.trial_executor.reset_trial(trial, {"hi": 1},
                                                 "modified_mock")
        self.assertEqual(exists, True)
        self.assertEqual(trial.config.get("hi"), 1)
        self.assertEqual(trial.experiment_tag, "modified_mock")
        self.assertEqual(Trial.RUNNING, trial.status) 

# 需要导入模块: from ray import tune [as 别名]
# 或者: from ray.tune import Trainable [as 别名]
def create_resettable_class():
    class MyResettableClass(Trainable):
        def setup(self, config):
            self.config = config
            self.num_resets = 0
            self.iter = 0

        def step(self):
            self.iter += 1
            return {"num_resets": self.num_resets, "done": self.iter > 1}

        def save_checkpoint(self, chkpt_dir):
            return {"iter": self.iter}

        def load_checkpoint(self, item):
            self.iter = item["iter"]

        def reset_config(self, new_config):
            if "fake_reset_not_supported" in self.config:
                return False
            self.num_resets += 1
            return True

    return MyResettableClass 

# 需要导入模块: from ray import tune [as 别名]
# 或者: from ray.tune import Trainable [as 别名]
def setUp(self):
        class MockTrainable(Trainable):
            scores_dict = {
                0: [5, 4, 4, 4, 4, 4, 4, 4, 0],
                1: [4, 3, 3, 3, 3, 3, 3, 3, 1],
                2: [2, 1, 1, 1, 1, 1, 1, 1, 8],
                3: [9, 7, 7, 7, 7, 7, 7, 7, 6],
                4: [7, 5, 5, 5, 5, 5, 5, 5, 3]
            }

            def setup(self, config):
                self.id = config["id"]
                self.idx = 0

            def step(self):
                val = self.scores_dict[self.id][self.idx]
                self.idx += 1
                return {"score": val}

            def save_checkpoint(self, checkpoint_dir):
                pass

            def load_checkpoint(self, checkpoint_path):
                pass

        self.MockTrainable = MockTrainable
        ray.init(local_mode=False, num_cpus=1) 

# 需要导入模块: from ray import tune [as 别名]
# 或者: from ray.tune import Trainable [as 别名]
def testCheckpointing(self):
        pbt = self.basicSetup(perturbation_interval=2)

        class train(tune.Trainable):
            def step(self):
                return {"mean_accuracy": self.training_iteration}

            def save_checkpoint(self, path):
                checkpoint = os.path.join(path, "checkpoint")
                with open(checkpoint, "w") as f:
                    f.write("OK")
                return checkpoint

        trial_hyperparams = {
            "float_factor": 2.0,
            "const_factor": 3,
            "int_factor": 10,
            "id_factor": 0
        }

        analysis = tune.run(
            train,
            num_samples=3,
            scheduler=pbt,
            checkpoint_freq=3,
            config=trial_hyperparams,
            stop={"training_iteration": 30})

        for trial in analysis.trials:
            self.assertEqual(trial.status, Trial.TERMINATED)
            self.assertTrue(trial.has_checkpoint()) 

# 需要导入模块: from ray import tune [as 别名]
# 或者: from ray.tune import Trainable [as 别名]
def testCheckpointDict(self):
        pbt = self.basicSetup(perturbation_interval=2)

        class train_dict(tune.Trainable):
            def setup(self, config):
                self.state = {"hi": 1}

            def step(self):
                return {"mean_accuracy": self.training_iteration}

            def save_checkpoint(self, path):
                return self.state

            def load_checkpoint(self, state):
                self.state = state

        trial_hyperparams = {
            "float_factor": 2.0,
            "const_factor": 3,
            "int_factor": 10,
            "id_factor": 0
        }

        analysis = tune.run(
            train_dict,
            num_samples=3,
            scheduler=pbt,
            checkpoint_freq=3,
            config=trial_hyperparams,
            stop={"training_iteration": 30})

        for trial in analysis.trials:
            self.assertEqual(trial.status, Trial.TERMINATED)
            self.assertTrue(trial.has_checkpoint()) 

# 需要导入模块: from ray import tune [as 别名]
# 或者: from ray.tune import Trainable [as 别名]
def TuneTrainable(train_fn):
    """Helper function for geting a trainable to use with Tune

    The function expectes a train_fn function which takes a config as input,
    and returns four items.

    - model: the tensorflow estimator.
    - train_spec: training specification.
    - eval_spec: evaluation specification.
    - reporter: a function which returns the metrics given evalution.

    The resulting trainable reports the metrics when checkpoints are saved,
    the report frequency is controlled by the checkpoint frequency,
    and the metrics are determined by reporter.
    """
    import os
    from ray.tune import Trainable
    from tensorflow.train import CheckpointSaverListener

    class _tuneStoper(CheckpointSaverListener):
        def after_save(self, session, global_step_value):
            return True

    class TuneTrainable(Trainable):
        def _setup(self, config):
            tf.logging.set_verbosity(tf.logging.ERROR)
            self.config = config
            model, train_spec, eval_spec, reporter = train_fn(config)
            self.model = model
            self.train_spec = train_spec
            self.eval_spec = eval_spec
            self.reporter = reporter

        def _train(self):
            import warnings
            index_warning = 'Converting sparse IndexedSlices'
            warnings.filterwarnings('ignore', index_warning)
            model = self.model
            model.train(input_fn=self.train_spec.input_fn,
                        max_steps=self.train_spec.max_steps,
                        hooks=self.train_spec.hooks,
                        saving_listeners=[_tuneStoper()])
            eval_out = model.evaluate(input_fn=self.eval_spec.input_fn,
                                      steps=self.eval_spec.steps,
                                      hooks=self.eval_spec.hooks)
            metrics = self.reporter(eval_out)
            return metrics

        def _save(self, checkpoint_dir):
            latest_checkpoint = self.model.latest_checkpoint()
            chkpath = os.path.join(checkpoint_dir, 'path.txt')
            with open(chkpath, 'w') as f:
                f.write(latest_checkpoint)
            return chkpath

        def _restore(self, checkpoint_path):
            with open(checkpoint_path) as f:
                chkpath = f.readline().strip()
            self.model, _, _, _ = train_fn(self.config, chkpath)
    return TuneTrainable