def __init__(
self,
model,
prefix,
input_record,
predict_input_record_fields=None,
tags=None,
**kwargs
):
"""
Base class for model layers. Layer is an abstraction that allows to
provide model description in terms of meta-operators, where each of the
meta-operators can have different implementations for training,
evaluation and prediction, that are instantiated later. As an example
SampledSoftmax can do something related to sampling depending on
supervision during the training and just apply softmax if it's used for
prediction/evaluation.
All inputs/outputs from layers are represented as a record (instance of
schema bounded to blobs) and are accessible through input_record and
output_schema. If Layer needs to have only a subset of inputs/provides
subset of outputs during the inference - it should provide
predict_input_record and predict_output_schema correspondingly (those
records are expected to be a subset of input_record/output_schema).
Each layer has a list of Tags associated with it, that depends on
current context and arguments. It's possible to use those tags during
the instantiation time.
"""
self.name = model.next_layer_name(prefix)
self.model = model
self.kwargs = kwargs
self._input_record = input_record
if predict_input_record_fields:
if not isinstance(predict_input_record_fields, list):
predict_input_record_fields = [predict_input_record_fields]
self._predict_input_record = self._input_record[predict_input_record_fields]
else:
self._predict_input_record = None
self.request_only = True
if len(input_record.all_scalars()) == 0:
self.request_only = False
for scalar in input_record.all_scalars():
if not is_request_only_scalar(scalar):
self.request_only = False
break
self.precomputation_request_only = False
self.precomputation_object_only = False
self._output_schema = None
self._predict_output_schema = None
self.eval_output_schema = None
self.tags = set(tags or [])
self.tags.update(TagContext.current().tags)
self.params = []
self._export_output_for_metrics = False
self._export_params_for_metrics = False
def __init__(self, model, prefix, input_record,
predict_input_record_fields=None, tags=None, **kwargs):
"""
Base class for model layers. Layer is an abstraction that allows to
provide model description in terms of meta-operators, where each of the
meta-operators can have different implementations for training,
evaluation and prediction, that are instantiated later. As an example
SampledSoftmax can do something related to sampling depending on
supervision during the training and just apply softmax if it's used for
prediction/evaluation.
All inputs/outputs from layers are represented as a record (instance of
schema bounded to blobs) and are accessible through input_record and
output_schema. If Layer needs to have only a subset of inputs/provides
subset of outputs during the inference - it should provide
predict_input_record and predict_output_schema correspondingly (those
records are expected to be a subset of input_record/output_schema).
Each layer has a list of Tags associated with it, that depends on
current context and arguments. It's possible to use those tags during
the instantiation time.
"""
self.name = model.next_layer_name(prefix)
self.model = model
self.kwargs = kwargs
self._input_record = input_record
if predict_input_record_fields:
if not isinstance(predict_input_record_fields, list):
predict_input_record_fields = [predict_input_record_fields]
self._predict_input_record = self._input_record[
predict_input_record_fields]
else:
self._predict_input_record = None
self.request_only = True
if len(input_record.all_scalars()) == 0:
self.request_only = False
for scalar in input_record.all_scalars():
if not is_request_only_scalar(scalar):
self.request_only = False
break
self._output_schema = None
self._predict_output_schema = None
self.eval_output_schema = None
self.tags = set(tags or [])
self.tags.update(TagContext.current().tags)
self.params = []
self._export_output_for_metrics = False
self._export_params_for_metrics = False
def __init__(self, model, prefix, input_record, tags=set(), **kwargs):
self.name = model.next_layer_name(prefix)
self.model = model
self.kwargs = kwargs
self.input_record = input_record
self.request_only = True
if len(input_record.all_scalars()) == 0:
self.request_only = False
for scalar in input_record.all_scalars():
if not _is_request_only_scalar(scalar):
self.request_only = False
break
self.output_schema = None
self.tags = set(tags)
self.tags.update(TagContext.current().tags)
self.params = []
def __init__(self, model, prefix, input_record, tags=set(), **kwargs):
self.name = model.next_block_name(prefix)
self.model = model
self.kwargs = kwargs
self.input_record = input_record
self.request_only = True
if len(input_record.all_scalars()) == 0:
self.request_only = False
for scalar in input_record.all_scalars():
if not _is_request_only_scalar(scalar):
self.request_only = False
break
self.output_schema = None
self.tags = set(tags)
self.tags.update(TagContext.current().tags)
self.params = []