def test_unaligned_metric_value(self):
modelBuilder_keras = KerasModelBuilder(model_creator_keras)
model = modelBuilder_keras.build(config={
"lr": 1e-2,
"batch_size": 32,
})
with pytest.raises(ValueError):
model.fit_eval(data=(self.data["x"], self.data["y"]),
validation_data=(self.data["val_x"], self.data["val_y"]),
metric='mae',
epochs=20)
def test_fit_evaluate(self):
modelBuilder_keras = KerasModelBuilder(model_creator_keras)
model = modelBuilder_keras.build(config={
"lr": 1e-2,
"batch_size": 32,
"metric": "mse"
})
val_result = model.fit_eval(data=(self.data["x"], self.data["y"]),
validation_data=(self.data["val_x"], self.data["val_y"]),
epochs=20)
assert val_result is not None
def test_uncompiled_model(self):
def model_creator(config):
"""Returns a tf.keras model"""
model = tf.keras.models.Sequential([
tf.keras.layers.Dense(1)
])
return model
modelBuilder_keras = KerasModelBuilder(model_creator)
with pytest.raises(ValueError):
model = modelBuilder_keras.build(config={
"lr": 1e-2,
"batch_size": 32,
"metric": "mse"
})
model.fit_eval(data=(self.data["x"], self.data["y"]),
validation_data=(self.data["val_x"], self.data["val_y"]),
epochs=20)
def from_keras(
*,
model_creator,
logs_dir="/tmp/auto_estimator_logs",
resources_per_trial=None,
name=None,
):
"""
Create an AutoEstimator for tensorflow keras.
:param model_creator: Tensorflow keras model creator function.
:param logs_dir: Local directory to save logs and results. It defaults to
"/tmp/auto_estimator_logs"
:param resources_per_trial: Dict. resources for each trial. e.g. {"cpu": 2}.
:param name: Name of the auto estimator.
:return: an AutoEstimator object.
"""
from zoo.automl.model import KerasModelBuilder
model_builder = KerasModelBuilder(model_creator=model_creator)
return AutoEstimator(model_builder=model_builder,
logs_dir=logs_dir,
resources_per_trial=resources_per_trial,
name=name)
# 2. you can also use the model builder with a fix config
model = modelBuilder.build(config={
"lr": 1e-2, # used in optimizer_creator
"batch_size": 32, # used in data_creator
})
model.fit_eval(data=data,
validation_data=validation_data,
epochs=20)
val_result_pytorch_manual = model.evaluate(x=validation_data[0],
y=validation_data[1],
metrics=['rmse'])
# 3. try another modelbuilder based on tfkeras
modelBuilder_keras = KerasModelBuilder(model_creator_keras)
model = modelBuilder_keras.build(config={
"lr": 1e-2, # used in optimizer_creator
"batch_size": 32, # used in data_creator
"metric": "mse"
})
model.fit_eval(data=data,
validation_data=validation_data,
epochs=20)
val_result_tensorflow_manual = model.evaluate(x=validation_data[0],
y=validation_data[1],
metrics=['rmse'])
print("Searched best model validation rmse:", searched_best_model)
print("Pytorch model validation rmse:", val_result_pytorch_manual)
print("Tensorflow model validation rmse:", val_result_tensorflow_manual)