def test_greedy_oracle_state_hypermodel_is_graph():
oracle = greedy.GreedyOracle(
hypermodel=utils.build_graph(),
objective='val_loss',
)
oracle.set_state(oracle.get_state())
assert isinstance(oracle.hypermodel, graph_module.Graph)
def test_greedy_oracle_populate_doesnt_crash_with_init_hps(get_best_trials):
hp = keras_tuner.HyperParameters()
tf.keras.backend.clear_session()
input_node = ak.ImageInput(shape=(32, 32, 3))
input_node.batch_size = 32
input_node.num_samples = 1000
output_node = ak.ImageBlock()(input_node)
head = ak.ClassificationHead(num_classes=10)
head.shape = (10, )
output_node = head(output_node)
graph = ak.graph.Graph(inputs=input_node, outputs=output_node)
graph.build(hp)
oracle = greedy.GreedyOracle(
initial_hps=task_specific.IMAGE_CLASSIFIER,
objective="val_loss",
seed=utils.SEED,
)
trial = mock.Mock()
trial.hyperparameters = hp
get_best_trials.return_value = [trial]
for i in range(10):
tf.keras.backend.clear_session()
values = oracle.populate_space("a")["values"]
hp = oracle.hyperparameters.copy()
hp.values = values
graph.build(hp)
oracle.update_space(hp)
def test_greedy_oracle_get_state_update_space_can_run():
oracle = greedy.GreedyOracle(
hypermodel=utils.build_graph(),
objective='val_loss',
)
oracle.set_state(oracle.get_state())
hp = kerastuner.HyperParameters()
hp.Boolean('test')
oracle.update_space(hp)
def test_random_oracle_state():
graph = utils.build_graph()
oracle = greedy.GreedyOracle(
hypermodel=graph,
objective='val_loss',
)
oracle.hypermodel = graph
oracle.set_state(oracle.get_state())
assert oracle.hypermodel is graph
def test_overwrite_search(fit_fn, base_tuner_search, tmp_path):
graph = utils.build_graph()
tuner = tuner_module.AutoTuner(
oracle=greedy.GreedyOracle(graph, objective='val_loss'),
hypermodel=graph,
directory=tmp_path)
tuner.search(epochs=10)
assert tuner._finished
def test_greedy_oracle_populate_space_with_no_hp(get_best_trials):
hp = keras_tuner.HyperParameters()
oracle = greedy.GreedyOracle(objective="val_loss", seed=utils.SEED)
trial = mock.Mock()
trial.hyperparameters = hp
get_best_trials.return_value = [trial]
oracle.update_space(hp)
values_a = oracle.populate_space("a")["values"]
assert len(values_a) == 0
def test_add_early_stopping(fit_fn, base_tuner_search, tmp_path):
graph = utils.build_graph()
tuner = tuner_module.AutoTuner(
oracle=greedy.GreedyOracle(graph, objective='val_loss'),
hypermodel=graph,
directory=tmp_path)
tuner.search(x=None, epochs=10)
callbacks = base_tuner_search.call_args_list[0][1]['callbacks']
assert any([isinstance(callback, tf.keras.callbacks.EarlyStopping)
for callback in callbacks])
def test_greedy_oracle_populate_different_values(get_best_trials):
hp = keras_tuner.HyperParameters()
utils.build_graph().build(hp)
oracle = greedy.GreedyOracle(objective="val_loss", seed=utils.SEED)
trial = mock.Mock()
trial.hyperparameters = hp
get_best_trials.return_value = [trial]
oracle.update_space(hp)
values_a = oracle.populate_space("a")["values"]
values_b = oracle.populate_space("b")["values"]
assert not all([values_a[key] == values_b[key] for key in values_a])
def test_no_epochs(best_epochs, fit_fn, base_tuner_search, tmp_path):
best_epochs.return_value = 2
graph = utils.build_graph()
tuner = tuner_module.AutoTuner(
oracle=greedy.GreedyOracle(graph, objective='val_loss'),
hypermodel=graph,
directory=tmp_path)
tuner.search(x=mock.Mock(), epochs=None, fit_on_val_data=True,
validation_data=mock.Mock())
callbacks = fit_fn.call_args_list[0][1]['callbacks']
print(callbacks)
assert not any([isinstance(callback, tf.keras.callbacks.EarlyStopping)
for callback in callbacks])
def test_greedy_oracle_stop_reach_max_collision(get_best_trials,
compute_values_hash):
hp = keras_tuner.HyperParameters()
utils.build_graph().build(hp)
oracle = greedy.GreedyOracle(objective="val_loss", seed=utils.SEED)
trial = mock.Mock()
trial.hyperparameters = hp
get_best_trials.return_value = [trial]
compute_values_hash.return_value = 1
oracle.update_space(hp)
oracle.populate_space("a")["values"]
assert (oracle.populate_space("b")["status"] ==
keras_tuner.engine.trial.TrialStatus.STOPPED)
def test_greedy_oracle(fn):
oracle = greedy.GreedyOracle(
hypermodel=utils.build_graph(),
objective='val_loss',
)
trial = mock.Mock()
hp = kerastuner.HyperParameters()
trial.hyperparameters = hp
fn.return_value = [trial]
oracle.update_space(hp)
for i in range(2000):
oracle._populate_space(str(i))
assert 'optimizer' in oracle._hp_names[greedy.GreedyOracle.OPT]
assert 'classification_head_1/dropout' in oracle._hp_names[
greedy.GreedyOracle.ARCH]
assert 'image_block_1/block_type' in oracle._hp_names[
greedy.GreedyOracle.HYPER]
def test_image_classifier_oracle():
tf.keras.backend.clear_session()
input_node = ak.ImageInput(shape=(32, 32, 3))
output_node = ak.ImageBlock()(input_node)
output_node = ak.ClassificationHead(loss='categorical_crossentropy',
output_shape=(10, ))(output_node)
graph = graph_module.Graph(input_node, output_node)
oracle = greedy.GreedyOracle(hypermodel=graph,
initial_hps=task_specific.IMAGE_CLASSIFIER,
objective='val_loss')
oracle._populate_space('0')
hp = oracle.get_space()
hp.values = task_specific.IMAGE_CLASSIFIER[0]
assert len(
set(task_specific.IMAGE_CLASSIFIER[0].keys()) -
set(oracle.get_space().values.keys())) == 0
oracle._populate_space('1')
assert len(
set(task_specific.IMAGE_CLASSIFIER[1].keys()) -
set(oracle.get_space().values.keys())) == 0
def test_greedy_oracle_populate_doesnt_crash_with_init_hps(get_best_trials):
hp = kerastuner.HyperParameters()
graph = utils.build_graph()
graph.build(hp)
oracle = greedy.GreedyOracle(
initial_hps=task_specific.IMAGE_CLASSIFIER,
objective="val_loss",
seed=utils.SEED,
)
trial = mock.Mock()
trial.hyperparameters = hp
get_best_trials.return_value = [trial]
for i in range(10):
tf.keras.backend.clear_session()
values = oracle._populate_space("a")["values"]
hp = oracle.hyperparameters.copy()
hp.values = values
graph.build(hp)
oracle.update_space(hp)
def test_greedy_oracle_get_state_update_space_can_run():
oracle = greedy.GreedyOracle(objective="val_loss")
oracle.set_state(oracle.get_state())
hp = keras_tuner.HyperParameters()
hp.Boolean("test")
oracle.update_space(hp)
