Skip to content

SinaAuto Api Documents

Jump to bottom
naili-xing edited this page Jul 25, 2020 · 13 revisions

SingaAuto Front-end / Back-end APIs

A. Dataset Upload Page

Existing APIs (with modification)

POST /datasets

Parameters:

  • Auth:
    • 'user_id'
  • File:
    • 'dataset': string
  • Form:
    • name: string
    • task: string

Response:

  • JSON
    • 'id': string
    • 'name': string
    • 'task': string
    • 'size_bytes': string

GET /datasets

Auth:

Parameters:

  • Auth:
    • 'user_id'

Response:

  • JSON
    • array:
      • 'id': dataset.id,
      • 'name': dataset.name,
      • 'task': dataset.task,
      • 'datetime_created': dataset.datetime_created,
      • 'size_bytes': dataset.size_bytes,
      • 'store_dataset_id':
      • 'stat'

B. Pre-train Model Selection Page

Existing APIs

POST /models

Parameters:

  • Auth:
    • user_id
  • Form:
    • name: string
    • task: string
    • model_file_bytes: file
    • model_class: string
    • checkpoint_id:string
    • docker_image=none
    • dependencies={}
    • access_right=ModelAccessRight.PRIVATE

Response:

  • 'id': model.id,
  • 'user_id': model.user_id,
  • 'name': model.name

DELETE /models/{model_id}

Parameters:

  • Auth:
    • user_id
  • Form:
    • model_id

Responses:

  • 'id': model.id,
  • 'user_id': model.user_id,
  • 'name': model.name

GET /models/{model_id}

Parameters:

  • Auth:
    • user_id

Responses:

  • 'id': model.id,
  • 'user_id': model.user_id,
  • 'name': model.name,
  • 'task': model.task,
  • 'model_class': model.model_class,
  • 'datetime_created': model.datetime_created,
  • 'docker_image': model.docker_image,
  • 'dependencies': model.dependencies,
  • 'access_right': model.access_right

GET /models/available?task=<your-task>

Parameters:

  • Auth:
    • user_id
  • Query param:
    • task=None

Responses:

  • array:
    • 'id': model.id,
    • 'user_id': model.user_id,
    • 'name': model.name,
    • 'task': model.task,
    • 'datetime_created': model.datetime_created,
    • 'dependencies': model.dependencies,
    • 'access_right': model.access_right

Proposed New APIs

GET /model/recommended

Parameters:

  • Auth
    • user_id
  • Form:
    • dataset_id

Response:

  • JSON
    • array
      • 'id': model.id,
      • 'user_id': model.user_id,
      • 'name': model.name,
      • 'task': model.task,
      • 'datetime_created': model.datetime_created,
      • 'dependencies': model.dependencies,
      • 'access_right': model.access_right

C. Training Status Page

Existing APIs

POST /train_jobs 

    # from admin.py
    def create_train_job(
        self, user_id, app, task,
        train_dataset_id, val_dataset_id,
        budget, model_ids, train_args={}):
        """
        Creates and starts a train job on Rafiki. 

        A train job is uniquely identified by user, its associated app, and the app version (returned in output).
        
        Only admins, model developers & app developers can manage train jobs. Model developers & app developers can only manage their own train jobs.

        :param app: Name of the app associated with the train job
        :param task: Task associated with the train job, 
            the train job will train models associated with the task
        :param train_dataset_id: ID of the train dataset, previously created on Rafiki
        :param val_dataset_id: ID of the validation dataset, previously created on Rafiki
        :param budget: Budget for train job
                The following describes the budget options available:

        =====================       =====================
        **Budget Option**             **Description**
        ---------------------       ---------------------
        ``TIME_HOURS``              Max no. of hours to train (soft target). Defaults to 0.1.
        ``GPU_COUNT``               No. of GPUs to allocate for training, across all models. Defaults to 0.
        ``MODEL_TRIAL_COUNT``       Max no. of trials to conduct for each model (soft target). -1 for unlimited. Defaults to -1.
        =====================       =====================
        ``budget`` should be a dictionary of ``{ <budget_type>: <budget_amount> }``, where 
        ``<budget_type>`` is one of :class:`rafiki.constants.BudgetOption` and 
        ``<budget_amount>`` specifies the amount for the associated budget option.
       
        :param model_ids: List of IDs of model to use for train job.
        NOTE: only client.py defaults to all models if model_ids is None!

        :param train_args: Additional arguments to pass to models during training, if any. 
            Refer to the task's specification for appropriate arguments  
        :returns: Created train job as dictionary
        """

Parameters:

  • Auth
    • user_id
  • Form
    • app,
    • task,
    • train_dataset_id,
    • val_dataset_id,
    • budget: string {},
    • model_ids: list in string quotation or list,
    • train_args={}

Responses:

  • 'id': train_job.id,
  • 'app': train_job.app,
  • 'app_version': train_job.app_version

GET /train_jobs

get /train_jobs is calling get_train_jobs_by_user

    # from admin.py
    def get_train_jobs_by_user(self, user_id):
        """
        unlike get_train_jobs_by_app,
        get_train_jobs_by_user is called by:
        @app.route('/train_jobs', methods=['GET'])
        """

Parameters: before Feb2020, /train_jobs expects user_id as a query-param

  • Auth:
    • user_id

Responses:

  • JSON:
    • array:
      • 'id': x.id,
      • 'status': x.status,
      • 'app': x.app,
      • 'app_version': x.app_version,
      • 'task': x.task,
      • 'train_dataset_id': x.train_dataset_id,
      • 'val_dataset_id': x.val_dataset_id,
      • 'train_args': x.train_args,
      • 'datetime_started': x.datetime_started,
      • 'datetime_stopped': x.datetime_stopped,
      • 'budget': x.budget

GET /train_jobs/{app} 

    # from admin.py
    def get_train_jobs_by_app(self, user_id, app):
        """
        unlike get_train_jobs_by_user,
        get_train_jobs_by_app is for:
        GET /train_jobs/{app}
        """

Parameters:

  • Auth:
    • user_id

Responses:

  • JSON:
    • array:
      • 'id': x.id,
      • 'status': x.status,
      • 'app': x.app,
      • 'app_version': x.app_version,
      • 'task': x.task,
      • 'train_dataset_id': x.train_dataset_id,
      • 'val_dataset_id': x.val_dataset_id,
      • 'train_args': x.train_args,
      • 'datetime_started': x.datetime_started,
      • 'datetime_stopped': x.datetime_stopped,
      • 'budget': x.budget

GET /train_jobs/{app}/{app_version=-1} 

    # from admin.py
    def get_train_job(self, user_id, app, app_version=-1): # by app ver
        """
        get_train_job() is called by:
        @app.route('/train_jobs/<app>/<app_version>',
        methods=['GET'])
        """

Parameters:

  • Auth:
    • user_id

Responses:

  • JSON
    • 'id': train_job.id,
    • 'status': train_job.status,
    • 'app': train_job.app,
    • 'app_version': train_job.app_version,
    • 'task': train_job.task,
    • 'train_dataset_id': train_job.train_dataset_id,
    • 'val_dataset_id': train_job.val_dataset_id,
    • 'train_args': train_job.train_args,
    • 'datetime_started': train_job.datetime_started,
    • 'datetime_stopped': train_job.datetime_stopped

POST /train_jobs/{app}/{app_version}/stop

stop_train_job

Parameters:

  • Auth:
    • user_id
  • File:
  • Form:

Responses:

  • 'id': train_job.id,
  • 'app': train_job.app,
  • 'app_version': train_job.app_version

D. Post-train Model Selection Page

Existing APIs

GET /train_jobs/{app}/{app_version}/trials

get_trials_of_train_job

Paramters:

  • Auth:
    • user_id
  • File:
  • Form:
    • max_count
  • InPath:
    • app
    • app_version

Responses:

  • 'id': trial.id,
  • 'proposal': trial.proposal,
  • 'datetime_started': trial.datetime_started,
  • 'status': trial.status,
  • 'datetime_stopped': trial.datetime_stopped,
  • 'model_name': model.name,
  • 'score': trial.score,
  • 'is_params_saved': trial.is_params_saved

GET /trials/{trial_id}

get_trial

Paramters:

  • Auth:
  • File:
  • Form:
  • InPath:
    • trial_id

Responses:

  • 'id': trial.id,
  • 'no': trial.no,
  • 'worker_id': trial.worker_id,
  • 'proposal': trial.proposal,
  • 'datetime_started': trial.datetime_started,
  • 'status': trial.status,
  • 'datetime_stopped': trial.datetime_stopped,
  • 'model_name': model.name,
  • 'score': trial.score,
  • 'is_params_saved': trial.is_params_saved

GET /trials/{trial_id}/logs

get_trial_logs

Paramters:

  • Auth:
  • File:
  • Form:
  • InPath:
    • trial_id

Responses:

  • 'plots': plots,
  • 'metrics': metrics,
  • 'messages': messages

GET /trials/{trial_id}/parameters

get_trial_parameters

Paramters:

  • Auth:
  • File:
  • Form:
  • InPath:
    • trial_id

Responses:

  • application/octet-stream (pickle dump)

E. Post-train Model Inference Page

Existing APIs

POST /inference_jobs

create_inference_job

Paramters:

  • Auth:
    • user_id
  • File:
  • Form:
    • app
    • app_version
    • budget
  • InPath:

Responses:

  • 'id': inference_job.id,
  • 'train_job_id': train_job.id,
  • 'app': train_job.app,
  • 'app_version': train_job.app_version,
  • 'predictor_host': predictor_service.host

GET /inference_jobs

get_inference_jobs_by_user

Paramters:

  • Auth:
    • user_id
    • user_type
  • File:
  • Form:
  • InPath:

Responses:

  • JSON:
    • array
      • 'id': inference_job.id,
      • 'status': inference_job.status,
      • 'train_job_id': train_job.id,
      • 'app': train_job.app,
      • 'app_version': train_job.app_version,
      • 'datetime_started': inference_job.datetime_started,
      • 'datetime_stopped': inference_job.datetime_stopped

GET /inference_jobs/{app}

get_inference_jobs_of_app

Paramters:

  • Auth:
    • user_id
  • File:
  • Form:
  • InPath:
    • app

Responses:

  • JSON:
    • array
      • 'id': inference_job.id,
      • 'status': inference_job.status,
      • 'train_job_id': train_job.id,
      • 'app': train_job.app,
      • 'app_version': train_job.app_version,
      • 'datetime_started': inference_job.datetime_started,
      • 'datetime_stopped': inference_job.datetime_stopped

GET /inference_jobs/{app}/{app_version}

get_running_inference_jobs

Paramters:

  • Auth:
    • user_id
  • File:
  • Form:
  • InPath:
    • app
    • app_version

Responses:

  • 'id': inference_job.id,
  • 'status': inference_job.status,
  • 'train_job_id': train_job.id,
  • 'app': train_job.app,
  • 'app_version': train_job.app_version,
  • 'datetime_started': inference_job.datetime_started,
  • 'datetime_stopped': inference_job.datetime_stopped,
  • 'predictor_host': predictor_service.host if predictor_service is not None else None

POST /inference_jobs/{app}/{app_version}/stop

stop_inference_job

Paramters:

  • Auth:
    • user_id
  • File:
  • Form:
  • InPath:
    • app
    • app_version

Responses:

  • 'id': inference_job.id,
  • 'train_job_id': train_job.id,
  • 'app': train_job.app,
  • 'app_version': train_job.app_version

Proposed New APIs

POST /inference_jobs/checkpoint

create_inference_job_by_checkpoint

Paramters:

  • Auth:
    • user_id
  • File:
  • Form:
    • model_name
    • budget
  • InPath:

Responses:

  • 'id': inference_job.id,
  • 'train_job_id': train_job.id,
  • 'app': train_job.app,
  • 'app_version': train_job.app_version,
  • 'predictor_host': predictor_service.host
Clone this wiki locally