Deployment Modules for AutoAI models¶

Web Service¶

For usage instruction, see Web Service.

class ibm_watsonx_ai.deployment.WebService(source_instance_credentials=None, source_project_id=None, source_space_id=None, target_instance_credentials=None, target_project_id=None, target_space_id=None, project_id=None, space_id=None, **kwargs)[source]¶

Bases: BaseDeployment

WebService is an Online Deployment class. With this class object, you can manage any online (WebService) deployment.

Parameters:

source_instance_credentials (dict) – credentials to the instance where the training was performed
source_project_id (str, optional) – ID of the Watson Studio project where the training was performed
source_space_id (str, optional) – ID of the Watson Studio Space where the training was performed
target_instance_credentials (dict) – credentials to the instance where you want to deploy
target_project_id (str, optional) – ID of the Watson Studio project where you want to deploy
target_space_id (str, optional) – ID of the Watson Studio Space where you want to deploy

create(model, deployment_name, serving_name=None, metadata=None, training_data=None, training_target=None, experiment_run_id=None, hardware_spec=None)[source]¶

Create a deployment from a model.

Parameters:

model (str) – name of the AutoAI model
deployment_name (str) – name of the deployment
training_data (pandas.DataFrame or numpy.ndarray, optional) – training data for the model
training_target (pandas.DataFrame or numpy.ndarray, optional) – target/label data for the model
serving_name (str, optional) – serving name of the deployment
metadata (dict, optional) – meta properties of the model
experiment_run_id (str, optional) – ID of a training/experiment (only applicable for AutoAI deployments)
hardware_spec (dict, optional) – hardware specification for the deployment

Example:

from ibm_watsonx_ai.deployment import WebService
from ibm_watsonx_ai import Credentials

deployment = WebService(
        source_instance_credentials=Credentials(...),
        source_project_id="...",
        target_space_id="...")

deployment.create(
       experiment_run_id="...",
       model=model,
       deployment_name='My new deployment',
       serving_name='my_new_deployment'
   )

delete(deployment_id=None)[source]¶

Delete a deployment.

Parameters:: deployment_id (str, optional) – ID of the deployment to be deleted, if empty, current deployment will be deleted

Example:

deployment = WebService(workspace=...)
# Delete current deployment
deployment.delete()
# Or delete a specific deployment
deployment.delete(deployment_id='...')

get(deployment_id)[source]¶

Get a deployment.

Parameters:: deployment_id (str) – ID of the deployment

Example:

deployment = WebService(workspace=...)
deployment.get(deployment_id="...")

get_params()[source]¶: Get deployment parameters.

list(limit=None)[source]¶

List deployments.

Parameters:: limit (int, optional) – set the limit for number of listed deployments, default is None (all deployments should be fetched)
Returns:: Pandas DataFrame with information about deployments
Return type:: pandas.DataFrame

Example:

deployment = WebService(workspace=...)
deployments_list = deployment.list()
print(deployments_list)

# Result:
#                  created_at  ...  status
# 0  2020-03-06T10:50:49.401Z  ...   ready
# 1  2020-03-06T13:16:09.789Z  ...   ready
# 4  2020-03-11T14:46:36.035Z  ...  failed
# 3  2020-03-11T14:49:55.052Z  ...  failed
# 2  2020-03-11T15:13:53.708Z  ...   ready

score(payload=Empty DataFrame Columns: [] Index: [], *, forecast_window=None, transaction_id=None)[source]¶

Online scoring. Payload is passed to the Service scoring endpoint where the model has been deployed.

Parameters:

payload (pandas.DataFrame or dict) – DataFrame with data to test the model or dictionary with keys observations and supporting_features, and DataFrames with data for observations and supporting_features to score forecasting models
forecast_window (int, optional) – size of forecast window, supported only for forcasting, supported for CPD 5.0 and later
transaction_id (str, optional) – ID under which the records should be saved in the payload table in IBM OpenScale

Returns:

dictionary with list of model output/predicted targets

Return type:

dict

Examples

predictions = web_service.score(payload=test_data)
print(predictions)

# Result:
# {'predictions':
#     [{
#         'fields': ['prediction', 'probability'],
#         'values': [['no', [0.9221385608558003, 0.07786143914419975]],
#                   ['no', [0.9798324002736079, 0.020167599726392187]]
#     }]}

predictions = web_service.score(payload={'observations': new_observations_df})
predictions = web_service.score(payload={'observations': new_observations_df, 'supporting_features': supporting_features_df}) # supporting features time series forecasting scenario
predictions = web_service.score(payload={'observations': new_observations_df}
                                forecast_window=1000) # forecast_window time series forecasting scenario

Batch¶

For usage instruction, see Batch.

class ibm_watsonx_ai.deployment.Batch(source_instance_credentials=None, source_project_id=None, source_space_id=None, target_instance_credentials=None, target_project_id=None, target_space_id=None, project_id=None, space_id=None, **kwargs)[source]¶

Bases: BaseDeployment

The Batch Deployment class. With this class object, you can manage any batch deployment.

Parameters:

source_instance_credentials (dict) – credentials to the instance where the training was performed
source_project_id (str, optional) – ID of the Watson Studio project where the training was performed
source_space_id (str, optional) – ID of the Watson Studio Space where the training was performed
target_instance_credentials (dict) – credentials to the instance where you want to deploy
target_project_id (str, optional) – ID of the Watson Studio project where you want to deploy
target_space_id (str, optional) – ID of the Watson Studio Space where you want to deploy

create(model, deployment_name, metadata=None, training_data=None, training_target=None, experiment_run_id=None, hardware_spec=None)[source]¶

Create a deployment from a model.

Parameters:

model (str) – name of the AutoAI model
deployment_name (str) – name of the deployment
training_data (pandas.DataFrame or numpy.ndarray, optional) – training data for the model
training_target (pandas.DataFrame or numpy.ndarray, optional) – target/label data for the model
metadata (dict, optional) – meta properties of the model
experiment_run_id (str, optional) – ID of a training/experiment (only applicable for AutoAI deployments)
hardware_spec (str, optional) – hardware specification name of the deployment

Example:

from ibm_watsonx_ai.deployment import Batch

deployment = Batch(
        source_instance_credentials=Credentials(...),
        source_project_id="...",
        target_space_id="...")

deployment.create(
       experiment_run_id="...",
       model=model,
       deployment_name='My new deployment'
       hardware_spec='L'
   )

delete(deployment_id=None)[source]¶

Delete a deployment.

Parameters:: deployment_id (str, optional) – ID of the deployment to be deleted, if empty, current deployment will be deleted

Example:

deployment = Batch(workspace=...)
# Delete current deployment
deployment.delete()
# Or delete a specific deployment
deployment.delete(deployment_id='...')

get(deployment_id)[source]¶

Get a deployment.

Parameters:: deployment_id (str) – ID of the deployment

Example:

deployment = Batch(workspace=...)
deployment.get(deployment_id="...")

get_job_id(batch_scoring_details)[source]¶: Get the ID from batch scoring details.

get_job_params(scoring_job_id=None)[source]¶

Get batch deployment job parameters.

Parameters:: scoring_job_id (str) – ID of the scoring job
Returns:: parameters of the scoring job
Return type:: dict

get_job_result(scoring_job_id)[source]¶

Get batch deployment results of a scoring job.

Parameters:: scoring_job_id (str) – ID of the scoring job
Returns:: batch deployment results of the scoring job
Return type:: pandas.DataFrame
Raises:: MissingScoringResults – in case of incompleted or failed job MissingScoringResults scoring exception is raised

get_job_status(scoring_job_id)[source]¶

Get the status of a scoring job.

Parameters:: scoring_job_id (str) – ID of the scoring job
Returns:: dictionary with state of the scoring job (one of: [completed, failed, starting, queued]) and additional details if they exist
Return type:: dict

get_params()[source]¶: Get deployment parameters.

list(limit=None)[source]¶

List deployments.

Parameters:: limit (int, optional) – set the limit for number of listed deployments, default is None (all deployments should be fetched)
Returns:: Pandas DataFrame with information about deployments
Return type:: pandas.DataFrame

Example:

deployment = Batch(workspace=...)
deployments_list = deployment.list()
print(deployments_list)

# Result:
#                  created_at  ...  status
# 0  2020-03-06T10:50:49.401Z  ...   ready
# 1  2020-03-06T13:16:09.789Z  ...   ready
# 4  2020-03-11T14:46:36.035Z  ...  failed
# 3  2020-03-11T14:49:55.052Z  ...  failed
# 2  2020-03-11T15:13:53.708Z  ...   ready

list_jobs()[source]¶: Returns pandas DataFrame with a list of deployment jobs

rerun_job(scoring_job_id, background_mode=True)[source]¶

Rerun scoring job with the same parameters as job described by scoring_job_id.

Parameters:

scoring_job_id (str) – ID of the described scoring job
background_mode (bool, optional) – indicator whether the score_rerun() method will run in the background (async) or (sync)

Returns:

details of the scoring job

Return type:

dict

Example:

scoring_details = deployment.score_rerun(scoring_job_id)

run_job(payload=Empty DataFrame Columns: [] Index: [], output_data_reference=None, transaction_id=None, background_mode=True, hardware_spec=None)[source]¶

Batch scoring job. Payload or Payload data reference is required. Passed to the Service where the model has been deployed.

Parameters:

payload (pandas.DataFrame or List[DataConnection] or Dict) – DataFrame that contains data to test the model or data storage connection details that inform the model where the payload data is stored
output_data_reference (DataConnection, optional) – DataConnection to the output COS for storing predictions, required only when DataConnections are used as a payload
transaction_id (str, optional) – ID under which the records should be saved in the payload table in IBM OpenScale
background_mode (bool, optional) – indicator whether the score() method will run in the background (async) or (sync)
hardware_spec (str, optional) – hardware specification name for the scoring job

Returns:

details of the scoring job

Return type:

dict

Examples

score_details = batch_service.run_job(payload=test_data)
print(score_details['entity']['scoring'])

# Result:
# {'input_data': [{'fields': ['sepal_length',
#               'sepal_width',
#               'petal_length',
#               'petal_width'],
#              'values': [[4.9, 3.0, 1.4, 0.2]]}],
# 'predictions': [{'fields': ['prediction', 'probability'],
#               'values': [['setosa',
#                 [0.9999320742502246,
#                  5.1519823540224506e-05,
#                  1.6405926235405522e-05]]]}]

payload_reference = DataConnection(location=DSLocation(asset_id=asset_id))
score_details = batch_service.run_job(payload=payload_reference, output_data_filename = "scoring_output.csv")
score_details = batch_service.run_job(payload={'observations': payload_reference})
score_details = batch_service.run_job(payload=[payload_reference])
score_details = batch_service.run_job(payload={'observations': payload_reference, 'supporting_features': supporting_features_reference})  # supporting features time series forecasting sceanrio
score_details = batch_service.run_job(payload=test_df, hardware_spec='S')
score_details = batch_service.run_job(payload=test_df, hardware_spec=TShirtSize.L)

score(**kwargs)[source]¶

Scoring on Service. Payload is passed to the scoring endpoint where the model has been deployed.

Parameters:: payload (pandas.DataFrame) – data to test the model