class DefaultPipeline:
"""The biomedicus default pipeline for processing clinical documents.
Attributes
events_client (mtap.EventsClient): An MTAP events client used by the pipeline.
pipeline (mtap.Pipeline): An MTAP pipeline to use to process documents.
"""
def __init__(self, conf: PipelineConf, *, events_client: EventsClient = None):
conf.populate_addresses()
if events_client is not None:
self.close_client = False
self.events_client = events_client
elif conf.events_address is not None:
self.close_client = True
self.events_client = EventsClient(address=conf.events_address)
else:
raise ValueError("Events client or address not specified.")
pipeline = [
(conf.sentences_id, conf.sentences_address),
(conf.section_headers_id, conf.section_headers_address),
(conf.tagger_id, conf.tagger_address),
(conf.acronyms_id, conf.acronyms_address),
(conf.concepts_id, conf.concepts_address),
(conf.negation_id, conf.negation_address),
(conf.selective_dependencies_id, conf.selective_dependencies_address),
(conf.deepen_id, conf.deepen_address)
]
if conf.use_discovery:
self.pipeline = Pipeline(
*[RemoteProcessor(identifier) for identifier, _ in pipeline]
)
else:
self.pipeline = Pipeline(
*[RemoteProcessor(identifier, address=addr) for identifier, addr in pipeline]
)
if conf.serializer is not None:
serialization_proc = SerializationProcessor(get_serializer(conf.serializer),
conf.output_directory,
include_label_text=conf.include_label_text)
ser_comp = LocalProcessor(serialization_proc, component_id='serializer',
client=self.events_client)
self.pipeline.append(ser_comp)
def process_text(self, text: str, *, event_id: str = None) -> ProcessingResult:
with Event(event_id=event_id, client=self.events_client) as event:
document = event.create_document('plaintext', text=text)
f = self.pipeline.run(document)
return f
def __enter__(self):
return self
def __exit__(self, exc_type, exc_val, exc_tb):
self.pipeline.close()
if self.close_client:
self.events_client.close()
class DefaultPipeline:
"""The biomedicus default pipeline for processing clinical documents.
Attributes
events_client (mtap.EventsClient): An MTAP events client used by the pipeline.
pipeline (mtap.Pipeline): An MTAP pipeline to use to process documents.
"""
def __init__(self,
conf_path: Union[str, Path],
output_directory: Union[str, Path],
*,
events_address: Optional[str] = None,
events_client: EventsClient = None,
serializer: Optional[str] = None,
include_label_text: bool = False):
if events_address == 'None' or events_address == 'none' or events_address == 'null' or events_address == '':
events_address = None
if events_client is not None:
self.close_client = False
self.events_client = events_client
else:
self.close_client = True
self.events_client = EventsClient(address=events_address)
self.pipeline = Pipeline.from_yaml_file(conf_path)
if serializer == 'None':
serializer = None
if serializer is not None:
serialization_proc = SerializationProcessor(
get_serializer(serializer),
output_directory,
include_label_text=include_label_text)
ser_comp = LocalProcessor(serialization_proc,
component_id='serializer',
client=self.events_client)
self.pipeline.append(ser_comp)
def process_text(self,
text: str,
*,
event_id: str = None) -> ProcessingResult:
with Event(event_id=event_id, client=self.events_client) as event:
document = event.create_document('plaintext', text=text)
f = self.pipeline.run(document)
return f
def __enter__(self):
return self
def __exit__(self, exc_type, exc_val, exc_tb):
self.pipeline.close()
if self.close_client:
self.events_client.close()
class DefaultPipeline:
def __init__(self,
conf: DefaultPipelineConf,
*,
events_client: EventsClient = None):
if events_client is not None:
self.close_client = False
self.events_client = events_client
elif conf.events_address is not None:
self.close_client = True
self.events_client = EventsClient(address=conf.events_address)
else:
raise ValueError("Events client or address not specified.")
pipeline = [(conf.sentences_id, conf.sentences_address),
(conf.tagger_id, conf.tagger_address),
(conf.acronyms_id, conf.acronyms_address),
(conf.concepts_id, conf.concepts_address),
(conf.negation_id, conf.negation_address)]
if conf.use_discovery:
self.pipeline = Pipeline(
*[RemoteProcessor(identifier) for identifier, _ in pipeline],
n_threads=conf.threads)
else:
self.pipeline = Pipeline(*[
RemoteProcessor(identifier, address=addr)
for identifier, addr in pipeline
],
n_threads=conf.threads)
if conf.serializer is not None:
serialization_proc = SerializationProcessor(
get_serializer(conf.serializer),
conf.output_directory,
include_label_text=conf.include_label_text)
ser_comp = LocalProcessor(serialization_proc,
component_id='serializer',
client=self.events_client)
self.pipeline.append(ser_comp)
def process_text(self,
text: str,
*,
event_id: str = None) -> ProcessingResult:
with Event(event_id=event_id, client=self.events_client) as event:
document = event.create_document('plaintext', text=text)
f = self.pipeline.run(document)
return f
def __enter__(self):
return self
def __exit__(self, exc_type, exc_val, exc_tb):
self.pipeline.close()
if self.close_client:
self.events_client.close()
class DefaultPipeline:
"""The biomedicus default pipeline for processing clinical documents.
Attributes
events_client (mtap.EventsClient): An MTAP events client used by the pipeline.
pipeline (mtap.Pipeline): An MTAP pipeline to use to process documents.
"""
def __init__(self,
conf: PipelineConf,
*,
events_client: EventsClient = None):
conf.populate_addresses()
if events_client is not None:
self.close_client = False
self.events_client = events_client
elif conf.events_address is not None:
self.close_client = True
self.events_client = EventsClient(address=conf.events_address)
else:
raise ValueError("Events client or address not specified.")
pipeline = [(conf.sentences_id, conf.sentences_address),
(conf.tagger_id, conf.tagger_address)]
if conf.use_discovery:
self.pipeline = Pipeline(
*[RemoteProcessor(identifier) for identifier, _ in pipeline])
else:
self.pipeline = Pipeline(*[
RemoteProcessor(identifier, address=addr)
for identifier, addr in pipeline
])
def process_text(self,
text: str,
*,
event_id: str = None) -> ProcessingResult:
with Event(event_id=event_id, client=self.events_client) as event:
document = event.create_document('plaintext', text=text)
f = self.pipeline.run(document)
return f
def __enter__(self):
return self
def __exit__(self, exc_type, exc_val, exc_tb):
self.pipeline.close()
if self.close_client:
self.events_client.close()
class Pipeline(MutableSequence['processing.ComponentDescriptor']):
"""An object which can be used to build and run a pipeline of remote and local processors.
Pipelines are a :obj:`~typing.MutableSequence` containing
one or more :obj:`~mtap.processing.pipeline.ComponentDescriptor`,
a pipeline can be modified after creation using this functionality.
Args:
*components (ComponentDescriptor):
A list of component descriptors created using :class:`RemoteProcessor` or
:class:`LocalProcessor`.
Keyword Args:
name (~typing.Optional[str]): An optional name for the pipeline, defaults to 'pipeline'.
config (~typing.Optional[Config]): An optional config override.
Examples:
Remote pipeline with name discovery:
>>> with mtap.Events() as events, mtap.Pipeline(
>>> RemoteProcessor('processor-1-id'),
>>> RemoteProcessor('processor-2-id'),
>>> RemoteProcessor('processor-3-id')
>>> ) as pipeline:
>>> for txt in txts:
>>> with events.open_event() as event:
>>> document = event.add_document('plaintext', txt)
>>> results = pipeline.run(document)
Remote pipeline using addresses:
>>> with mtap.Events(address='localhost:50051') as events, mtap.Pipeline(
>>> RemoteProcessor('processor-1-name', address='localhost:50052'),
>>> RemoteProcessor('processor-2-id', address='localhost:50053'),
>>> RemoteProcessor('processor-3-id', address='localhost:50054')
>>> ) as pipeline:
>>> for txt in txts:
>>> event = events.open_event()
>>> document = event.add_document('plaintext', txt)
>>> results = pipeline.run(document)
Modifying pipeline
>>> pipeline = Pipeline(RemoteProcessor('foo', address='localhost:50000'),
RemoteProcessor('bar', address='localhost:50000'))
>>> pipeline
Pipeline(RemoteProcessor(processor_id='foo', address='localhost:50000', component_id=None, params=None),
RemoteProcessor(processor_id='bar', address='localhost:50000', component_id=None, params=None))
>>> pipeline.append(RemoteProcessor('baz', address='localhost:50001'))
>>> pipeline
Pipeline(RemoteProcessor(processor_id='foo', address='localhost:50000', component_id=None, params=None),
RemoteProcessor(processor_id='bar', address='localhost:50000', component_id=None, params=None),
RemoteProcessor(processor_id='baz', address='localhost:50001', component_id=None, params=None))
>>> del pipeline[1]
>>> pipeline
Pipeline(RemoteProcessor(processor_id='foo', address='localhost:50000', component_id=None, params=None),
RemoteProcessor(processor_id='baz', address='localhost:50001', component_id=None, params=None))
>>> pipeline[1] = RemoteProcessor(processor_id='bar', address='localhost:50003')
>>> pipeline
Pipeline(RemoteProcessor(processor_id='foo', address='localhost:50000', component_id=None, params=None),
RemoteProcessor(processor_id='bar', address='localhost:50003', component_id=None, params=None))
>>> pipeline += list(pipeline) # Putting in a new list to prevent an infinite recursion
>>> pipeline
Pipeline(RemoteProcessor(processor_id='foo', address='localhost:50000', component_id=None, params=None),
RemoteProcessor(processor_id='bar', address='localhost:50003', component_id=None, params=None),
RemoteProcessor(processor_id='foo', address='localhost:50000', component_id=None, params=None),
RemoteProcessor(processor_id='bar', address='localhost:50003', component_id=None, params=None))
Attributes:
name (str): The pipeline's name.
"""
__slots__ = [
'_component_ids', 'name', '_component_descriptors', 'events_address',
'mp_config', '_created_events_client', '_events_client', 'times_map',
'__components'
]
def __init__(self,
*components: 'processing.ComponentDescriptor',
name: Optional[str] = None,
events_address: Optional[str] = None,
events_client: Optional[EventsClient] = None,
mp_config: Optional[MpConfig] = None):
self._component_ids = {}
self.name = name or 'pipeline'
self._component_descriptors = list(components)
self.events_address = events_address
self._created_events_client = False
self._events_client = None
if events_client is not None:
self.events_client = events_client
self.mp_config = mp_config or MpConfig()
self.times_map = {}
def __reduce__(self):
return _create_pipeline, (self.name, self.events_address,
self._events_client, self.mp_config) + tuple(
self._component_descriptors)
@staticmethod
def from_yaml_file(conf_path: Union[pathlib.Path, str]) -> 'Pipeline':
"""Creates a pipeline from a yaml pipeline configuration file.
Args:
conf_path (str or pathlib.Path): The path to the configuration file.
Returns:
Pipeline object from the configuration.
"""
conf_path = pathlib.Path(conf_path)
from yaml import load
try:
from yaml import CLoader as Loader
except ImportError:
from yaml import Loader
with conf_path.open('rb') as f:
conf = load(f, Loader=Loader)
return Pipeline.load_configuration(conf)
@staticmethod
def load_configuration(conf: Dict) -> 'Pipeline':
"""Creates a pipeline from a pipeline configuration dictionary.
Args:
conf (Dict): The pipeline configuration dictionary.
Returns:
Pipeline created from the configuration.
"""
name = conf.get('name', None)
events_address = conf.get('events_address', None) or conf.get(
'events_addresses', None)
components = []
conf_components = conf.get('components', [])
for conf_component in conf_components:
components.append(
RemoteProcessor(processor_id=conf_component['processor_id'],
address=conf_component['address'],
component_id=conf_component.get(
'component_id', None),
params=dict(conf_component.get('params', {}))))
mp_config = MpConfig.from_configuration(conf.get('mp_config', {}))
return Pipeline(*components,
name=name,
events_address=events_address,
mp_config=mp_config)
@property
def events_client(self) -> EventsClient:
if self._events_client is not None:
return self._events_client
self._created_events_client = True
self._events_client = EventsClient(address=self.events_address)
return self._events_client
@events_client.setter
def events_client(self, value: EventsClient):
self._events_client = value
@property
def _components(self) -> 'List[processing.ProcessingComponent]':
try:
return self.__components
except AttributeError:
self.__components = [
desc.create_pipeline_component(self._component_ids,
lambda: self.events_client)
for desc in self._component_descriptors
]
return self.__components
@_components.deleter
def _components(self):
for component in self.__components:
component.close()
del self.__components
def run_multithread(self,
source: Union[Iterable[Union['mtap.Document',
'mtap.Event']],
'processing.ProcessingSource'],
*,
params: Optional[Dict[str, Any]] = None,
show_progress: Optional[bool] = None,
total: Optional[int] = None,
close_events: Optional[bool] = None,
max_failures: Optional[int] = None,
workers: Optional[int] = None,
read_ahead: Optional[int] = None,
mp_context=None):
"""Runs this pipeline on a source which provides multiple documents / events.
Concurrency is per-event, with each event being provided a thread which runs it through the
pipeline.
Args:
source (~typing.Union[~typing.Iterable[~typing.Union[Event, Document]], ProcessingSource])
A generator of events or documents to process. This should be an
:obj:`~typing.Iterable` of either :obj:`Event` or :obj:`Document` objects or a
:obj:`~mtap.processing.ProcessingSource`.
params (~typing.Optional[dict[str, ~typing.Any]])
Json object containing params specific to processing this event, the existing params
dictionary defined in :func:`~PipelineBuilder.add_processor` will be updated with
the contents of this dict.
show_progress (~typing.Optional[bool])
Whether to print a progress bar using tqdm.
total (~typing.Optional[int])
An optional argument indicating the total number of events / documents that will be
provided by the iterable, for the progress bar.
close_events (~typing.Optional[bool])
Whether the pipeline should close events after they have been fully processed
through all components.
max_failures (~typing.Optional[int])
The number of acceptable failures. Once this amount is exceeded processing will
halt. Note that because of the nature of conccurrency processing may continue for a
short amount of time before termination.
workers (~typing.Optional[int])
The number of threads to process documents on.
read_ahead (~typing.Optional[int])
The number of source documents to read ahead into memory before processing.
mp_context (multiprocessing context, optional)
An optional override for the multiprocessing context.
Examples:
>>> docs = list(Path('abc/').glob('*.txt'))
>>> def document_source():
>>> for path in docs:
>>> with path.open('r') as f:
>>> txt = f.read()
>>> with Event(event_id=path.name, client=client) as event:
>>> doc = event.create_document('plaintext', txt)
>>> yield doc
>>>
>>> pipeline.run_multithread(document_source(), total=len(docs))
"""
show_progress = show_progress if show_progress is not None else self.mp_config.show_progress
close_events = close_events if close_events is not None else self.mp_config.close_events
max_failures = max_failures if max_failures is not None else self.mp_config.max_failures
workers = workers if workers is not None else self.mp_config.workers
mp_context = (multiprocessing.get_context(
self.mp_config.mp_start_method)
if mp_context is None else mp_context)
read_ahead = read_ahead if read_ahead is not None else self.mp_config.read_ahead
with _PipelineMultiRunner(self, source, params, show_progress, total,
close_events, max_failures, workers,
read_ahead, mp_context) as runner:
runner.run()
def run(
self,
target: Union['mtap.Event', 'mtap.Document'],
*,
params: Optional[Dict[str,
Any]] = None) -> 'processing.PipelineResult':
"""Processes the event/document using all of the processors in the pipeline.
Args:
target (~typing.Union[Event, Document]): Either an event or a document to process.
params (dict[str, ~typing.Any]):
Json object containing params specific to processing this event, the existing params
dictionary defined in :func:`~PipelineBuilder.add_processor` will be updated with
the contents of this dict.
Returns:
list[ProcessingResult]: The results of all the processors in the pipeline.
Examples:
>>> e = mtap.Event()
>>> document = mtap.Document('plaintext', text="...", event=e)
>>> with Pipeline(...) as pipeline:
>>> pipeline.run(document)
>>> # is equivalent to pipeline.run(document.event, params={'document_name': document.document_name})
The 'document_name' param is used to indicate to :obj:`~mtap.DocumentProcessor`
which document on the event to process.
"""
event, params = _event_and_params(target, params)
event_id = event.event_id
result = self._run_by_event_id(event_id,
event.event_service_instance_id, params)
self._add_result_times(result)
for component_result in result.component_results:
try:
event.add_created_indices(component_result.created_indices)
except AttributeError:
pass
return result
def _run_by_event_id(self, event_id, event_service_instance_id, params):
start = datetime.now()
results = [
component.call_process(event_id, event_service_instance_id, params)
for component in self._components
]
total = datetime.now() - start
results = [
_base.ProcessingResult(identifier=component.component_id,
result_dict=result[0],
timing_info=result[1],
created_indices=result[2])
for component, result in zip(self._components, results)
]
logger.debug('Finished processing event_id: %s', event_id)
return _base.PipelineResult(results, total)
def _add_result_times(self, result):
times = {}
for component_id, _, component_times, _ in result.component_results:
times.update({
component_id + ':' + k: v
for k, v in component_times.items()
})
times[self.name + 'total'] = result.elapsed_time
_timing.add_times(self.times_map, times)
@overload
def processor_timer_stats(self) -> 'List[processing.AggregateTimingInfo]':
"""Returns the timing information for all processors.
Returns:
List[AggregateTimingInfo]:
A list of timing info objects, one for each processor, in the same order
that the processors were added to the pipeline.
"""
...
@overload
def processor_timer_stats(
self, identifier: str) -> 'processing.AggregateTimingInfo':
"""Returns the timing info for one processor.
Args:
identifier (Optional[str]): The pipeline component_id for the processor to return
timing info.
Returns:
AggregateTimingInfo: The timing info for the specified processor.
"""
...
def processor_timer_stats(self, identifier=None):
if identifier is not None:
aggregates = _timing.create_timer_stats(self.times_map,
identifier + ':')
aggregates = {
k[(len(identifier) + 1):]: v
for k, v in aggregates.items()
}
return _base.AggregateTimingInfo(identifier=identifier,
timing_info=aggregates)
timing_infos = []
for component in self._components:
component_id = component.component_id
aggregates = _timing.create_timer_stats(self.times_map,
component_id + ':')
aggregates = {
k[(len(component_id) + 1):]: v
for k, v in aggregates.items()
}
timing_infos.append(
_base.AggregateTimingInfo(identifier=component_id,
timing_info=aggregates))
return timing_infos
def pipeline_timer_stats(self) -> 'processing.AggregateTimingInfo':
"""The aggregated statistics for the global runtime of the pipeline.
Returns:
AggregateTimingInfo: The timing stats for the global runtime of the pipeline.
"""
pipeline_id = self.name
aggregates = _timing.create_timer_stats(self.times_map, pipeline_id)
aggregates = {k[len(pipeline_id):]: v for k, v in aggregates.items()}
return _base.AggregateTimingInfo(identifier=self.name,
timing_info=aggregates)
def __enter__(self):
return self
def __exit__(self, exc_type, exc_val, exc_tb):
self.close()
def close(self):
"""Closes any open connections to remote processors.
"""
for component in self._components:
try:
component.close()
except AttributeError:
pass
if self._created_events_client:
self._events_client.close()
def as_processor(self) -> 'processing.EventProcessor':
"""Returns the pipeline as a processor.
Returns:
EventProcessor: An event processor that can be added to other pipelines or hosted.
"""
return _PipelineProcessor(self._components)
def print_times(self):
"""Prints all of the times collected during this pipeline using :func:`print`.
"""
self.pipeline_timer_stats().print_times()
for pipeline_timer in self.processor_timer_stats():
pipeline_timer.print_times()
def __getitem__(self, item):
return self._component_descriptors[item]
def __setitem__(self, key, value):
self._clear_components()
self._component_descriptors[key] = value
def __delitem__(self, key):
self._clear_components()
del self._component_descriptors[key]
def __len__(self):
return len(self._component_descriptors)
def _clear_components(self):
try:
del self._components
except AttributeError:
pass
def insert(self, index, o) -> None:
self._clear_components()
self._component_descriptors.insert(index, o)
def __repr__(self):
return "Pipeline(" + ', '.join(
[repr(component)
for component in self._component_descriptors]) + ')'