def _test_base_tuner():
def build_model(hp):
return hp.Int("a", 1, 100)
tuner = SimpleTuner(
oracle=kt.oracles.RandomSearch(objective=kt.Objective(
"score", "max"),
max_trials=10),
hypermodel=build_model,
directory=tmp_dir,
)
tuner.search()
# Only worker makes it to this point, server runs until thread stops.
assert dist_utils.has_chief_oracle()
assert not dist_utils.is_chief_oracle()
assert isinstance(tuner.oracle,
kt.distribute.oracle_client.OracleClient)
barrier.wait(60)
# Model is just a score.
scores = tuner.get_best_models(10)
assert len(scores)
assert scores == sorted(copy.copy(scores), reverse=True)
def __init__(
self,
oracle,
hypermodel=None,
directory=None,
project_name=None,
logger=None,
overwrite=False,
):
# Ops and metadata
self.directory = directory or "."
self.project_name = project_name or "untitled_project"
if overwrite and tf.io.gfile.exists(self.project_dir):
tf.io.gfile.rmtree(self.project_dir)
if not isinstance(oracle, oracle_module.Oracle):
raise ValueError(
"Expected `oracle` argument to be an instance of `Oracle`. "
f"Received: oracle={oracle} (of type ({type(oracle)}).")
self.oracle = oracle
self.oracle._set_project_dir(self.directory,
self.project_name,
overwrite=overwrite)
# Run in distributed mode.
if dist_utils.is_chief_oracle():
# Blocks forever.
oracle_chief.start_server(self.oracle)
elif dist_utils.has_chief_oracle():
# Proxies requests to the chief oracle.
self.oracle = oracle_client.OracleClient(self.oracle)
# To support tuning distribution.
self.tuner_id = os.environ.get("KERASTUNER_TUNER_ID", "tuner0")
self.hypermodel = hm_module.get_hypermodel(hypermodel)
# Logs etc
self.logger = logger
self._display = tuner_utils.Display(oracle=self.oracle)
self._populate_initial_space()
if not overwrite and tf.io.gfile.exists(self._get_tuner_fname()):
tf.get_logger().info("Reloading Tuner from {}".format(
self._get_tuner_fname()))
self.reload()
def _test_random_search():
def build_model(hp):
model = keras.Sequential()
model.add(keras.layers.Dense(3, input_shape=(5, )))
for i in range(hp.Int("num_layers", 1, 3)):
model.add(
keras.layers.Dense(hp.Int("num_units_%i" % i, 1, 3),
activation="relu"))
model.add(keras.layers.Dense(1, activation="sigmoid"))
model.compile("sgd", "binary_crossentropy")
return model
x = np.random.uniform(-1, 1, size=(2, 5))
y = np.ones((2, 1))
tuner = kt.tuners.RandomSearch(
hypermodel=build_model,
objective="val_loss",
max_trials=10,
directory=tmp_dir,
)
# Only worker makes it to this point, server runs until thread stops.
assert dist_utils.has_chief_oracle()
assert not dist_utils.is_chief_oracle()
assert isinstance(tuner.oracle,
kt.distribute.oracle_client.OracleClient)
tuner.search(x, y, validation_data=(x, y), epochs=1, batch_size=2)
# Suppress warnings about optimizer state not being restored by tf.keras.
tf.get_logger().setLevel(logging.ERROR)
trials = tuner.oracle.get_best_trials(2)
assert trials[0].score <= trials[1].score
models = tuner.get_best_models(2)
assert models[0].evaluate(x, y) <= models[1].evaluate(x, y)