def test_fetch_job(turing_api, active_project, api_response_get, expected,
api_response_refresh, updated, use_google_oauth):
turing.set_url(turing_api, use_google_oauth)
turing.set_project(active_project.name)
responses.add(method="GET",
url=f"/v1/projects/{active_project.id}/jobs/{expected.id}",
body=api_response_get,
status=200,
content_type="application/json")
job = turing.batch.EnsemblingJob.get_by_id(expected.id)
assert job == expected
responses.reset()
responses.add(method="GET",
url=f"/v1/projects/{active_project.id}/jobs/{expected.id}",
body=api_response_refresh,
status=200,
content_type="application/json")
job.refresh()
assert job == updated
def test_submit_job(turing_api, active_project, ensembling_job_config,
api_response, expected, use_google_oauth):
turing.set_url(turing_api, use_google_oauth)
turing.set_project(active_project.name)
responses.add(method="POST",
url=f"/v1/projects/{active_project.id}/jobs",
body=api_response,
status=201,
content_type="application/json")
actual = turing.batch.job.EnsemblingJob.submit(
ensembler_id=2,
config=ensembling_job_config,
)
assert actual == expected
def main(turing_api: str, project: str):
# Initialize Turing client
turing.set_url(turing_api)
turing.set_project(project)
# List projects
projects = turing.Project.list()
for p in projects:
print(p)
# Save pyfunc ensembler in Turing's backend
ensembler = turing.PyFuncEnsembler.create(
name="my-ensembler",
ensembler_instance=MyEnsembler(),
conda_env={
'dependencies': [
'python>=3.8.0',
# other dependencies, if required
]
})
print("Ensembler created:\n", ensembler)
# Update Ensembler's name
ensembler.update(name="my-ensembler-updated")
print("Updated:\n", ensembler)
# Update Ensembler's implementation
ensembler.update(
ensembler_instance=MyEnsembler(),
conda_env={
'channels': ['defaults'],
'dependencies': ['python=3.7.0', "cookiecutter>=1.7.2", "numpy"]
},
code_dir=[os.path.join(os.path.dirname(__file__), "../../samples")],
)
print("Updated:\n", ensembler)
# List pyfunc ensemblers
ensemblers = turing.PyFuncEnsembler.list()
for e in ensemblers:
print(e)
def test_terminate_job(turing_api, active_project, job, api_response_delete,
api_response_get, expected, use_google_oauth):
turing.set_url(turing_api, use_google_oauth)
turing.set_project(active_project.name)
responses.add(method="DELETE",
url=f"/v1/projects/{active_project.id}/jobs/{job.id}",
body=api_response_delete,
status=201,
content_type="application/json")
responses.add(method="GET",
url=f"/v1/projects/{active_project.id}/jobs/{job.id}",
body=api_response_get,
status=200,
content_type="application/json")
assert job != expected
job.terminate()
assert job == expected
def test_list_jobs(turing_api, active_project, api_response, expected,
use_google_oauth):
turing.set_url(turing_api, use_google_oauth)
turing.set_project(active_project.name)
responses.add(method="GET",
url=f"/v1/projects/{active_project.id}/jobs?"
f"status={turing.batch.EnsemblingJobStatus.PENDING.value}&"
f"status={turing.batch.EnsemblingJobStatus.RUNNING.value}",
body=api_response,
match_querystring=True,
status=200,
content_type="application/json")
actual = turing.batch.EnsemblingJob.list(status=[
turing.batch.EnsemblingJobStatus.PENDING,
turing.batch.EnsemblingJobStatus.RUNNING
])
assert len(actual) == len(expected)
for actual, expected in zip(actual, expected):
assert actual == expected
def main(turing_api: str, project: str):
# Initialize Turing client:
turing.set_url(turing_api)
turing.set_project(project)
# Save pyfunc ensembler in Turing's backend:
ensembler = turing.PyFuncEnsembler.create(
name="my-ensembler",
ensembler_instance=MyEnsembler(),
conda_env={
'dependencies': [
'python>=3.8.0',
# other dependencies, if required
]
})
print("Ensembler created:\n", ensembler)
# Or fetch existing ensembler by its ID:
# ensembler_id = < ENSEMBLER_ID >
# ensembler = turing.PyFuncEnsembler.get_by_id(ensembler_id)
# Define configuration of the batch ensembling job
# Configure datasource, that contains input features:
source = turing.batch.config.source.BigQueryDataset(
table="project.dataset.features",
features=["feature_1", "feature_2",
"features_3"]).join_on(columns=["feature_1"])
# Configure dataset(s), that contain predictions of individual models:
predictions = {
'model_odd':
turing.batch.config.source.BigQueryDataset(
table="project.dataset.scores_model_odd",
features=["feature_1", "prediction_score"]).join_on(
columns=["feature_1"]).select(columns=["prediction_score"]),
'model_even':
turing.batch.config.source.BigQueryDataset(
query="""
SELECT feature_1, prediction_score
FROM `project.dataset.scores_model_even`
WHERE target_date = DATE("2021-03-15", "Asia/Jakarta")
""",
options={
"viewsEnabled": "true",
"materializationDataset": "my_dataset"
}).join_on(columns=["feature_1"]).select(
columns=["prediction_score"])
}
# Configure ensembling result:
result_config = turing.batch.config.ResultConfig(
type=turing.batch.config.ResultType.INTEGER,
column_name="prediction_result")
# Configure destination, where ensembling results will be stored:
sink = turing.batch.config.sink.BigQuerySink(
table="project.dataset.ensembling_results",
staging_bucket="staging_bucket"
).save_mode(turing.batch.config.sink.SaveMode.OVERWRITE) \
.select(columns=["feature_1", "feature_2", "prediction_result"])
# (Optional) Configure resources allocation for the job execution:
resource_request = turing.batch.config.ResourceRequest(
driver_cpu_request="1",
driver_memory_request="1G",
executor_replica=5,
executor_cpu_request="500Mi",
executor_memory_request="800M")
# (Optional) Configure environment variables here
env_vars = {
"SOME_VAR": "SOME_VALUE",
}
# Submit the job for execution:
job = ensembler.submit_job(
turing.batch.config.EnsemblingJobConfig(
source=source,
predictions=predictions,
result_config=result_config,
sink=sink,
service_account=SERVICE_ACCOUNT_NAME,
resource_request=resource_request,
env_vars=env_vars,
))
print(job)
# You can also retrieve the instance of existing job by its ID:
# job_id = < JOB_ID >
# job = turing.batch.EnsemblingJob.get_by_id(job_id=job_id)
#
# # Or list all ensembling jobs within the project
# jobs = turing.batch.EnsemblingJob.list(status=[
# turing.batch.EnsemblingJobStatus.PENDING,
# turing.batch.EnsemblingJobStatus.RUNNING,
# ])
# Refresh the status of the job
for i in range(3):
time.sleep(5)
job.refresh()
print(f"Refresh #{i+1}: {job}")
# It's also possible to terminate a running job:
job.terminate()
print(f"Job's termination in process: {job}")