aup.EE.Experiment module

aup.EE.Experiment is called by aup.__main__ to start an experiment.

See Configure HPO Algorithm for configuration specification.

APIs

class Experiment(exp_config, username=None, connector=None, auppath='.aup', sleep_time=1, eid=None, start=True, request_stop_time=5)

Bases: object

Experiment Class - create and run an experiment

Parameters

• exp_config (BasicConfig.BasicConfig) – configuration of the experiment

• username (str) – username, default: None - will use login username

• connector (AbstractConnector) – connector to database

• auppath (str) – Auptimizer env.ini file folder, default is either ./.aup or ~/.aup

• sleep_time (int) – time to pause between jobs

add_refresh_signal()

add_suspend_signal()

finish()

Finish experiment if no job is running

Returns

job id, best score

Return type

(int, float)

resume(filename)

Restore previous experiment, previous job during suspension won’t be run in this round

Parameters

filename (str) – filename (saved by pickle as exp%d.pkl)

start()

Start experiment

submit_job(job=None, rid_blacklist=None)

Submit a new job to run if there is resource available :param job: optional job parameter in case a job needs resubmitting :type job: aup.EE.Job.Job object :param rid_blacklist: resource ids to exclude when submitting job :type rid_blacklist: [int]

Returns

True if job submitted, else False

update(score, jid)

Callback function passed to aup.EE.Resource.AbstractResourceManager to

update the job history (also proposer and connector)

Parameters

• score (float) – score returned from job (using aup.utils.print_result())

• jid (int) – job id