def get_relevance_model(self,
feature_layer_keys_to_fns={}) -> RelevanceModel:
"""
Creates RelevanceModel suited for classification use-case.
NOTE: Override this method to create custom loss, scorer, model objects.
"""
# Define interaction model
interaction_model: InteractionModel = UnivariateInteractionModel(
feature_config=self.feature_config,
feature_layer_keys_to_fns=feature_layer_keys_to_fns,
tfrecord_type=self.tfrecord_type,
file_io=self.file_io)
# Define loss object from loss key
loss: RelevanceLossBase = categorical_cross_entropy.get_loss(
loss_key=self.loss_key)
# Define scorer
scorer: ScorerBase = RelevanceScorer.from_model_config_file(
model_config_file=self.model_config_file,
interaction_model=interaction_model,
loss=loss,
output_name=self.args.output_name,
logger=self.logger,
file_io=self.file_io,
)
# Define metrics objects from metrics keys
metrics: List[Union[Type[Metric], str]] = [
metric_factory.get_metric(metric_key=metric_key)
for metric_key in self.metrics_keys
]
# Define optimizer
optimizer: Optimizer = get_optimizer(
optimizer_key=self.optimizer_key,
learning_rate=self.args.learning_rate,
learning_rate_decay=self.args.learning_rate_decay,
learning_rate_decay_steps=self.args.learning_rate_decay_steps,
gradient_clip_value=self.args.gradient_clip_value,
)
# Combine the above to define a RelevanceModel
relevance_model: RelevanceModel = RelevanceModel(
feature_config=self.feature_config,
scorer=scorer,
metrics=metrics,
optimizer=optimizer,
tfrecord_type=self.tfrecord_type,
model_file=self.args.model_file,
compile_keras_model=self.args.compile_keras_model,
output_name=self.args.output_name,
file_io=self.local_io,
logger=self.logger,
)
return relevance_model
def from_univariate_interaction_model(
cls,
model_config,
feature_config: FeatureConfig,
tfrecord_type: str,
loss: RelevanceLossBase,
metrics: List[Union[kmetrics.Metric, str]],
optimizer: Optimizer,
feature_layer_keys_to_fns: dict = {},
model_file: Optional[str] = None,
compile_keras_model: bool = False,
output_name: str = "score",
max_sequence_size: int = 0,
file_io: FileIO = None,
logger=None,
):
"""Use this as constructor to use UnivariateInteractionModel and RelevanceScorer"""
interaction_model: InteractionModel = UnivariateInteractionModel(
feature_config=feature_config,
feature_layer_keys_to_fns=feature_layer_keys_to_fns,
tfrecord_type=tfrecord_type,
max_sequence_size=max_sequence_size,
)
return cls.from_relevance_scorer(
interaction_model=interaction_model,
model_config=model_config,
feature_config=feature_config,
loss=loss,
metrics=metrics,
optimizer=optimizer,
tfrecord_type=tfrecord_type,
model_file=model_file,
compile_keras_model=compile_keras_model,
output_name=output_name,
file_io=file_io,
logger=logger,
)
def get_relevance_model(self, feature_layer_keys_to_fns={}) -> RelevanceModel:
"""
Creates a RelevanceModel that can be used for training and evaluating
Parameters
----------
feature_layer_keys_to_fns : dict of (str, function)
dictionary of function names mapped to tensorflow compatible
function definitions that can now be used in the InteractionModel
as a feature function to transform input features
Returns
-------
`RelevanceModel`
RelevanceModel that can be used for training and evaluating
a classification model
Notes
-----
Override this method to create custom loss, scorer, model objects
"""
# Define interaction model
interaction_model: InteractionModel = UnivariateInteractionModel(
feature_config=self.feature_config,
feature_layer_keys_to_fns=feature_layer_keys_to_fns,
tfrecord_type=self.tfrecord_type,
file_io=self.file_io,
)
# Define loss object from loss key
loss: RelevanceLossBase = categorical_cross_entropy.get_loss(loss_key=self.loss_key)
# Define scorer
scorer: ScorerBase = RelevanceScorer(
feature_config=self.feature_config,
model_config=self.model_config,
interaction_model=interaction_model,
loss=loss,
output_name=self.args.output_name,
logger=self.logger,
file_io=self.file_io,
)
# Define metrics objects from metrics keys
metrics: List[Union[Type[Metric], str]] = [
metrics_factory.get_metric(metric_key=metric_key) for metric_key in self.metrics_keys
]
# Define optimizer
optimizer: Optimizer = get_optimizer(model_config=self.model_config)
# Combine the above to define a RelevanceModel
relevance_model: RelevanceModel = ClassificationModel(
feature_config=self.feature_config,
scorer=scorer,
metrics=metrics,
optimizer=optimizer,
tfrecord_type=self.tfrecord_type,
model_file=self.args.model_file,
initialize_layers_dict=ast.literal_eval(self.args.initialize_layers_dict),
freeze_layers_list=ast.literal_eval(self.args.freeze_layers_list),
compile_keras_model=self.args.compile_keras_model,
output_name=self.args.output_name,
file_io=self.local_io,
logger=self.logger,
)
return relevance_model
def test_cyclic_lr_in_training_pipeline(self):
"""Test a cyclic learning rate in model training"""
Logger = logging_utils.setup_logging(
reset=True,
file_name=os.path.join(INPUT_DIR + 'ranklib', "output_log.csv"),
log_to_file=True,
)
io = LocalIO()
feature_config = self.parse_config(
TFRecordTypeKey.SEQUENCE_EXAMPLE,
self.feature_config_yaml_convert_to_clicks, io)
dataset = RelevanceDataset(
data_dir=INPUT_DIR + '/ranklib',
data_format=DataFormatKey.RANKLIB,
feature_config=feature_config,
tfrecord_type=TFRecordTypeKey.SEQUENCE_EXAMPLE,
batch_size=2,
file_io=io,
preprocessing_keys_to_fns={},
logger=Logger,
keep_additional_info=KEEP_ADDITIONAL_INFO,
non_zero_features_only=NON_ZERO_FEATURES_ONLY,
max_sequence_size=319,
)
# Define interaction model
interaction_model: InteractionModel = UnivariateInteractionModel(
feature_config=feature_config,
feature_layer_keys_to_fns={},
tfrecord_type=TFRecordTypeKey.SEQUENCE_EXAMPLE,
max_sequence_size=319,
file_io=io,
)
# Define loss object from loss key
loss: RelevanceLossBase = loss_factory.get_loss(
loss_key=LossKey.RANK_ONE_LISTNET,
scoring_type=ScoringTypeKey.POINTWISE)
# Define scorer
scorer: ScorerBase = RelevanceScorer.from_model_config_file(
model_config_file=self.model_config_file,
interaction_model=interaction_model,
loss=loss,
logger=Logger,
file_io=io,
)
optimizer: Optimizer = get_optimizer(
model_config=io.read_yaml(self.model_config_file))
# Combine the above to define a RelevanceModel
relevance_model: RelevanceModel = RankingModel(
feature_config=feature_config,
tfrecord_type=TFRecordTypeKey.SEQUENCE_EXAMPLE,
scorer=scorer,
optimizer=optimizer,
model_file=None,
file_io=io,
logger=Logger,
)
callbacks_list = []
my_callback_object = LrCallback()
callbacks_list.append(my_callback_object)
history = relevance_model.model.fit(
x=dataset.train,
validation_data=dataset.validation,
epochs=2,
verbose=True,
callbacks=callbacks_list,
)
lr_list = my_callback_object.get_lr_list()
lr_gold = [
0.001, 0.020800006, 0.040599994, 0.0604, 0.080199994, 0.1,
0.080199994, 0.0604, 0.040599994, 0.020800006, 0.001, 0.010900003,
0.020800006, 0.030699994, 0.040599994, 0.050499998, 0.040599994,
0.030699994, 0.020800006, 0.010900003, 0.001, 0.0059499955,
0.010900003, 0.015849996, 0.020800006, 0.02575, 0.020800006,
0.015849996, 0.010900003, 0.0059499955, 0.001, 0.0034749978,
0.0059500015, 0.008424998, 0.010900003, 0.013375, 0.010900003,
0.008424998, 0.0059500015, 0.0034749978, 0.001, 0.0022374988,
0.0034749978, 0.0047125025, 0.0059500015, 0.0071875, 0.0059500015,
0.0047125025
]
for i in range(len(lr_list)):
assert np.isclose(lr_gold[i], lr_list[i])
def from_univariate_interaction_model(
cls,
model_config,
feature_config: FeatureConfig,
tfrecord_type: str,
loss: RelevanceLossBase,
metrics: List[Union[kmetrics.Metric, str]],
optimizer: Optimizer,
feature_layer_keys_to_fns: dict = {},
model_file: Optional[str] = None,
initialize_layers_dict: dict = {},
freeze_layers_list: list = [],
compile_keras_model: bool = False,
output_name: str = "score",
max_sequence_size: int = 0,
file_io: FileIO = None,
logger=None,
):
"""
Create a RelevanceModel with default UnivariateInteractionModel
Parameters
----------
feature_config : `FeatureConfig` object
FeatureConfig object that defines the features to be loaded in the dataset
and the preprocessing functions to be applied to each of them
model_config : dict
dictionary defining the dense model architecture
tfrecord_type : {"example", "sequence_example"}
Type of the TFRecord protobuf message used for TFRecordDataset
file_io : `FileIO` object
file I/O handler objects for reading and writing data
loss : `RelevanceLossBase` object
Loss object defining the final activation layer and the loss function
metrics : list
List of keras Metric classes that will be used for evaluating the trained model
optimizer : `Optimizer`
Tensorflow keras optimizer to be used for training the model
feature_layer_keys_to_fns : dict
Dictionary of custom feature transformation functions to be applied
on the input features as part of the InteractionModel
model_file : str, optional
Path to pretrained model file to be loaded for evaluation or retraining
initialize_layers_dict : dict, optional
Dictionary of tensorflow layer names mapped to the path of pretrained weights
Use this for transfer learning with pretrained weights
freeze_layers_list : list, optional
List of model layer names to be frozen
Use this for freezing pretrained weights from other ml4ir models
compile_keras_model : bool, optional
Whether the keras model loaded from disk should be compiled
with loss, metrics and an optimizer
output_name : str, optional
Name of the output tensorflow node that captures the score
max_sequence_size : int, optional
Maximum length of the sequence to be used for SequenceExample protobuf objects
logger : `Logger`, optional
logging handler for status messages
Returns
-------
RelevanceModel
RelevanceModel object with a UnivariateInteractionModel
"""
interaction_model: InteractionModel = UnivariateInteractionModel(
feature_config=feature_config,
feature_layer_keys_to_fns=feature_layer_keys_to_fns,
tfrecord_type=tfrecord_type,
max_sequence_size=max_sequence_size,
)
return cls.from_relevance_scorer(
interaction_model=interaction_model,
model_config=model_config,
feature_config=feature_config,
loss=loss,
metrics=metrics,
optimizer=optimizer,
tfrecord_type=tfrecord_type,
model_file=model_file,
initialize_layers_dict=initialize_layers_dict,
freeze_layers_list=freeze_layers_list,
compile_keras_model=compile_keras_model,
output_name=output_name,
file_io=file_io,
logger=logger,
)
def get_ranking_model(
self,
loss_key: str,
metrics_keys: List,
feature_config: FeatureConfig,
feature_layer_keys_to_fns={},
) -> RelevanceModel:
"""
Creates RankingModel
NOTE: Override this method to create custom loss, scorer, model objects
"""
# Define interaction model
interaction_model: InteractionModel = UnivariateInteractionModel(
feature_config=feature_config,
feature_layer_keys_to_fns=feature_layer_keys_to_fns,
tfrecord_type=self.args.tfrecord_type,
max_sequence_size=self.args.max_sequence_size,
)
# Define loss object from loss key
loss: RelevanceLossBase = loss_factory.get_loss(
loss_key=loss_key, scoring_type=self.args.scoring_type)
# Define scorer
scorer: ScorerBase = RelevanceScorer.from_model_config_file(
model_config_file=self.args.model_config,
interaction_model=interaction_model,
loss=loss,
output_name=self.args.output_name,
)
# Define metrics objects from metrics keys
metrics: List[Union[Type[Metric], str]] = [
metric_factory.get_metric(metric_key=metric_key)
for metric_key in metrics_keys
]
# Define optimizer
optimizer: Optimizer = get_optimizer(
optimizer_key=self.args.optimizer_key,
learning_rate=self.args.learning_rate,
learning_rate_decay=self.args.learning_rate_decay,
learning_rate_decay_steps=self.args.learning_rate_decay_steps,
gradient_clip_value=self.args.gradient_clip_value,
)
# Combine the above to define a RelevanceModel
relevance_model: RelevanceModel = RankingModel(
feature_config=feature_config,
tfrecord_type=self.args.tfrecord_type,
scorer=scorer,
metrics=metrics,
optimizer=optimizer,
model_file=self.args.model_file,
compile_keras_model=self.args.compile_keras_model,
output_name=self.args.output_name,
logger=self.logger,
)
return relevance_model
def test_reduce_lr_on_plateau_in_training_pipeline(self):
"""Test reduce lr on plateau"""
self.model_config_file = MODEL_CONFIG_REDUCE_LR_ON_PLATEAU
Logger = logging_utils.setup_logging(
reset=True,
file_name=os.path.join(INPUT_DIR + 'ranklib', "output_log.csv"),
log_to_file=True,
)
io = LocalIO()
feature_config = self.parse_config(TFRecordTypeKey.SEQUENCE_EXAMPLE, self.feature_config_yaml_convert_to_clicks,
io)
model_config = io.read_yaml(self.model_config_file)
dataset = RelevanceDataset(
data_dir=INPUT_DIR + '/ranklib',
data_format=DataFormatKey.RANKLIB,
feature_config=feature_config,
tfrecord_type=TFRecordTypeKey.SEQUENCE_EXAMPLE,
batch_size=32,
file_io=io,
preprocessing_keys_to_fns={},
logger=Logger,
keep_additional_info=KEEP_ADDITIONAL_INFO,
non_zero_features_only=NON_ZERO_FEATURES_ONLY,
max_sequence_size=319,
)
# Define interaction model
interaction_model: InteractionModel = UnivariateInteractionModel(
feature_config=feature_config,
feature_layer_keys_to_fns={},
tfrecord_type=TFRecordTypeKey.SEQUENCE_EXAMPLE,
max_sequence_size=319,
file_io=io,
)
# Define loss object from loss key
loss: RelevanceLossBase = loss_factory.get_loss(
loss_key=LossKey.RANK_ONE_LISTNET, scoring_type=ScoringTypeKey.POINTWISE
)
# Define scorer
scorer: ScorerBase = RelevanceScorer.from_model_config_file(
model_config_file=self.model_config_file,
interaction_model=interaction_model,
loss=loss,
logger=Logger,
file_io=io,
)
optimizer: Optimizer = get_optimizer(model_config=model_config)
# Combine the above to define a RelevanceModel
relevance_model: RelevanceModel = RankingModel(
feature_config=feature_config,
tfrecord_type=TFRecordTypeKey.SEQUENCE_EXAMPLE,
scorer=scorer,
optimizer=optimizer,
model_file=None,
file_io=io,
logger=Logger,
)
callback_list = []
callback_list.append(relevance_model.define_scheduler_as_callback(None, model_config))
my_callback_object = LrCallback()
callback_list.append(my_callback_object)
history = relevance_model.model.fit(
x=dataset.train.shard(2, 0),
validation_data=dataset.validation.shard(2, 1),
epochs=10,
verbose=True,
callbacks=callback_list,
)
lr_list = my_callback_object.get_lr_reduce_on_plateau_list()
lr_gold = [50.0, 50.0, 25.0, 12.5, 6.25, 3.125, 1.5625, 1.0, 1.0, 1.0]
assert np.all(np.isclose(lr_gold, lr_list))
def get_ranking_model(
self,
loss_key: str,
metrics_keys: List,
feature_config: FeatureConfig,
model_config: dict = {},
feature_layer_keys_to_fns={},
initialize_layers_dict={},
freeze_layers_list=[],
) -> RelevanceModel:
"""
Creates RankingModel
NOTE: Override this method to create custom loss, scorer, model objects
"""
# Define interaction model
interaction_model: InteractionModel = UnivariateInteractionModel(
feature_config=feature_config,
feature_layer_keys_to_fns=feature_layer_keys_to_fns,
tfrecord_type=self.args.tfrecord_type,
max_sequence_size=self.args.max_sequence_size,
file_io=self.file_io,
)
# Define loss object from loss key
loss: RelevanceLossBase = loss_factory.get_loss(
loss_key=loss_key, scoring_type=self.args.scoring_type)
# Define scorer
scorer: ScorerBase = RelevanceScorer(
feature_config=feature_config,
model_config=self.model_config,
interaction_model=interaction_model,
loss=loss,
output_name=self.args.output_name,
logger=self.logger,
file_io=self.file_io,
)
# Define metrics objects from metrics keys
metrics: List[Union[Type[Metric], str]] = [
metric_factory.get_metric(metric_key=metric_key)
for metric_key in metrics_keys
]
# Define optimizer
optimizer: Optimizer = get_optimizer(
model_config=self.file_io.read_yaml(self.args.model_config), )
# Combine the above to define a RelevanceModel
if self.model_config["architecture_key"] == ArchitectureKey.LINEAR:
RankingModelClass = LinearRankingModel
else:
RankingModelClass = RankingModel
relevance_model: RelevanceModel = RankingModelClass(
feature_config=feature_config,
tfrecord_type=self.args.tfrecord_type,
scorer=scorer,
metrics=metrics,
optimizer=optimizer,
model_file=self.args.model_file,
initialize_layers_dict=initialize_layers_dict,
freeze_layers_list=freeze_layers_list,
compile_keras_model=self.args.compile_keras_model,
output_name=self.args.output_name,
logger=self.logger,
file_io=self.file_io,
)
return relevance_model