def GetCreatePipelineResults(self, request, context):
"""Mock GetCreatePipelineResults response"""
sessioncontext = request.context
if not sessioncontext.session_id in self.sessions:
yield core_pb2.PipelineCreateResult(\
response_info=core_pb2.Response(\
status=core_pb2.Status(\
code=core_pb2.FAILED_PRECONDITION,
details="Unknown session id: %s" % sessioncontext.session_id)))
return
results = [
(core_pb2.COMPLETED, 'pipeline_1', True),
(core_pb2.COMPLETED, 'pipeline_2', True),
]
cnt = 0
for progress, pipeline_id, send_pipeline in results:
#print('sleep 1 second...')
#time.sleep(1)
if not context.is_active():
logger.info("Client closed GetCreatePipelineResults stream")
msg = core_pb2.PipelineCreateResult(
response_info=core_pb2.Response(
status=core_pb2.Status(code=core_pb2.OK), ),
progress_info=progress,
pipeline_id=pipeline_id,
)
if send_pipeline:
cnt += 1
# try to create a legit file uri
file_uri_dict = get_predict_file_info_dict('CLASSIFICATION')
msg.pipeline_info.CopyFrom(
core_pb2.Pipeline(
#predict_result_uris=['file:///out/predict1.csv'],
predict_result_uri = file_uri_dict.get(\
TEST_KEY_FILE_URI,
'no file uri'),
output=core_pb2.OUTPUT_TYPE_UNDEFINED,
scores=[
core_pb2.Score(
metric=core_pb2.ACCURACY,
value=0.8,
),
core_pb2.Score(
metric=core_pb2.ROC_AUC,
value=0.5,
),
],
)
)
yield msg
def RunningPipeline(self, pipeline):
response = self._create_response("Pipeline Running")
progress = self._create_progress("RUNNING")
self.session.update_pipeline(pipeline)
result = core.PipelineCreateResult(
response_info = response,
progress_info = progress,
pipeline_id = pipeline.id
)
self.session.cache_planner_result(pipeline, result)
return result
def SubmittedPipeline(self, pipeline):
response = self._create_response("Pipeline Submitted")
progress = self._create_progress("SUBMITTED")
self.session.add_pipeline(pipeline)
result = core.PipelineCreateResult(
response_info = response,
progress_info = progress,
pipeline_id = pipeline.id
)
self.session.cache_planner_result(pipeline, result)
return result
def _response_session_invalid(self, session_id):
"Returns a message that the given session does not exist"
pipeline = core_pb2.Pipeline(predict_result_uri="invalid",
output=core_pb2.OUTPUT_TYPE_UNDEFINED,
scores=[])
msg = core_pb2.PipelineCreateResult(response_info=core_pb2.Response(
status=core_pb2.Status(code=core_pb2.SESSION_UNKNOWN), ),
progress_info=core_pb2.ERRORED,
pipeline_id="invalid",
pipeline_info=pipeline)
return msg
def CompletedPipeline(self, pipeline, exec_pipeline):
response = self._create_response("Pipeline Completed", "OK")
progress = self._create_progress("COMPLETED")
pipeline_info = None
if exec_pipeline is None:
response = self._create_response("Pipeline Failed", "INTERNAL")
else:
pipeline_info = self._create_pipeline_info(exec_pipeline)
# Update session pipeline
self.session.update_pipeline(exec_pipeline)
result = core.PipelineCreateResult(
response_info = response,
progress_info = progress,
pipeline_id = pipeline.id,
pipeline_info = pipeline_info
)
self.session.cache_planner_result(pipeline, result)
return result
def CreatePipelines(self, request, context):
"""Mock CreatePipelines response"""
sessioncontext = request.context
if not sessioncontext.session_id in self.sessions:
yield core_pb2.PipelineCreateResult(\
response_info=core_pb2.Response(\
status=core_pb2.Status(\
code=core_pb2.FAILED_PRECONDITION,
details="Unknown session id: %s" % sessioncontext.session_id)))
return
dataset_uri = request.dataset_uri
task = request.task
task_subtype = request.task_subtype
task_description = request.task_description
output = request.output
metrics = request.metrics
target_features = request.target_features
predict_features = request.predict_features
max_pipelines = request.max_pipelines
logger.info("Got CreatePipelines request, session=%s",
sessioncontext.session_id)
results = [
(core_pb2.SUBMITTED, 'pipeline_1', False),
(core_pb2.SUBMITTED, 'pipeline_2', False),
(core_pb2.RUNNING, 'pipeline_2', False),
(core_pb2.RUNNING, 'pipeline_1', False),
(core_pb2.COMPLETED, 'pipeline_1', True),
(core_pb2.COMPLETED, 'pipeline_2', True),
]
cnt = 0
for progress, pipeline_id, send_pipeline in results:
print('sleep 1 second...')
time.sleep(1)
if not context.is_active():
logger.info("Client closed CreatePipelines stream")
msg = core_pb2.PipelineCreateResult(
response_info=core_pb2.Response(
status=core_pb2.Status(code=core_pb2.OK), ),
progress_info=progress,
pipeline_id=pipeline_id,
)
if send_pipeline:
cnt += 1
# try to create a legit file uri
file_uri_dict = get_predict_file_info_dict('CLASSIFICATION')
msg.pipeline_info.CopyFrom(
core_pb2.Pipeline(
#predict_result_uris=['file:///out/predict1.csv'],
predict_result_uri=file_uri_dict.get(\
TEST_KEY_FILE_URI,
'no file uri'),
output=output,
scores=[
core_pb2.Score(
metric=core_pb2.ACCURACY,
value=0.8,
),
core_pb2.Score(
metric=core_pb2.ROC_AUC,
value=0.5,
),
],
)
)
yield msg
def ProblemNotImplemented(self):
response = self._create_response("Not Implemented", "UNIMPLEMENTED")
result = core.PipelineCreateResult(
response_info = response
)
return result
def CreatePipelines(self, request, context):
logging.info("Message received: CreatePipelines: %s", request)
session_id = request.context.session_id
if session_id not in self._sessions:
logging.warning(
"Asked to create pipeline for session %s which does not exist",
session_id)
return self._response_session_invalid(session_id)
session = self._sessions[session_id]
# Setup pipeline specification
dataset_uri = request.dataset_uri
task_type = request.task
# TODO: task_subtype is currently ignored.
# TODO: task_description is currently ignored.
metrics = request.metrics
target_features = request.target_features
predict_features = request.predict_features
# We need to tell the TA1 where it can find output,
# which will be in a created subdirectory of the dataset URI
# This assumes the URI is always file:// but changing that will
# be Hard so it's an ok assumption for now.
dataset_directory = dataset_uri_path(request.dataset_uri)
output_directory = os.path.join(dataset_directory, "output")
# We describe a set of related pipelines with a ProblemDescription.
# This basically is all the parameters for the problem that the client
# wants us to solve, with the idea that it will do whatever metalearning
# stuff it wants and then produce a set of PipelineDescription,
# where each Pipeline is a particular attempt at solving that problem.
spec = problem.ProblemDescription(session_id, dataset_uri,
output_directory, task_type, metrics,
target_features, predict_features)
logging.info("Starting new problem for session %s", session_id)
problem_id = session.new_problem(spec)
pipelines = spec.find_solutions(dataset_uri)
for pipeline in pipelines:
pipeline_id = self._new_pipeline_id()
output_file = pipeline_id + ".csv"
output_uri = "file://" + output_directory + "/" + output_file
pb2_pipeline = core_pb2.Pipeline(
predict_result_uri=output_uri,
output=core_pb2.OUTPUT_TYPE_UNDEFINED,
scores=[])
msg = core_pb2.PipelineCreateResult(
response_info=core_pb2.Response(
status=core_pb2.Status(code=core_pb2.OK), ),
progress_info=core_pb2.SUBMITTED,
pipeline_id=pipeline_id,
pipeline_info=pb2_pipeline)
yield msg
# Actually ask the problem description to start finding solutions.
msg.progress_info = core_pb2.RUNNING
pipeline.train(dataset_uri)
yield msg
pipeline.evaluate(dataset_uri,
os.path.join(output_directory, output_file),
target_features)
msg.progress_info = core_pb2.COMPLETED
yield msg