Welcome to watson-machine-learning-client’s documentation!¶

watson-machine-learning-client is a python library that allows to work with Watson Machine Learning service on IBM Cloud. Test and deploy your models as APIs for application development, share with colleagues using this python library.

NOTE!! DEPRECATED!! This Watson Machine Learning client version is deprecated starting Sep 1st, 2020 and will be discontinued at the end of the migration period. Refer to the documentation at https://dataplatform.cloud.ibm.com/docs/content/wsj/analyze-data/wml-ai.html for the migration process to be able to access new features

NOTE: DEPRECATED!! Python 3.6 framework is deprecated and will be removed on Jan 20th, 2021. It will be read-only mode starting Nov 20th, 2020. i.e you won’t be able to create new assets using this client. For cloud, switch to using IBM Watson Machine Learning Python client with python 3.7

NOTE: DEPRECATED!! Spark 2.3 framework is deprecated and will be removed on Dec 1st, 2020. For cloud, switch to using IBM Watson Machine Learning Python client with Spark 2.4

Contents

Welcome to watson-machine-learning-client’s documentation!
Indices and tables

Installation ¶

The package is available on pypi. Please use below command to install it.

$pip install watson-machine-learning-client

Note

This package is being replaced in favor of the new package “ibm-watson-machine-learning” in V4 released version of WML python client.

Requirements ¶

Watson Machine Learning service instance, you can create instance using this link. Only Python 3.5 or newer is supported.

Supported machine learning frameworks ¶

For the list of supported machine learning frameworks (models) please refer to Watson Machine Learning Documentation.

Sample notebooks ¶

Sample notebooks created using watson machine learning python client are available here.

API ¶

Authorization can be done by IAM token.

IAM token

from watson_machine_learning_client import WatsonMachineLearningAPIClient

wml_credentials = {
                   "url": "https://us-south.ml.cloud.ibm.com",
                   "apikey":"***********",
                   "instance_id": "*****"
                  }

client = WatsonMachineLearningAPIClient(wml_credentials)

service_instance ¶

class client.ServiceInstance(client)[source]¶

Connect, get details and check usage of your Watson Machine Learning service instance.

get_api_key()[source]¶

Get api_key of Watson Machine Learning service.

Returns: api_key
Return type: str

Example:

>>> instance_details = client.service_instance.get_api_key()

get_details()[source]¶

Get information about our instance of Watson Machine Learning service.

Returns: metadata of service instance
Return type: dict

Example:

>>> instance_details = client.service_instance.get_details()

get_instance_id()[source]¶

Get instance id of Watson Machine Learning service.

Returns: instance id
Return type: str

Example:

>>> instance_details = client.service_instance.get_instance_id()

get_password()[source]¶

Get password for Watson Machine Learning service.

Returns: password
Return type: str

Example:

>>> instance_details = client.service_instance.get_password()

get_url()[source]¶

Get instance url of Watson Machine Learning service.

Returns: instance url
Return type: str

Example:

>>> instance_details = client.service_instance.get_url()

get_username()[source]¶

Get username for Watson Machine Learning service.

Returns: username
Return type: str

Example:

>>> instance_details = client.service_instance.get_username()

repository ¶

class client.Repository(client)[source]¶

Store and manage your models, definitions and experiments using Watson Machine Learning Repository.

DefinitionMetaNames = <watson_machine_learning_client.metanames.ModelDefinitionMetaNames object>¶: MetaNames for definitions creation.

ExperimentMetaNames = <watson_machine_learning_client.metanames.ExperimentMetaNames object>¶: MetaNames for experiments creation.

FunctionMetaNames = <watson_machine_learning_client.metanames.FunctionMetaNames object>¶: MetaNames for python functions creation.

ModelMetaNames = <watson_machine_learning_client.metanames.ModelMetaNames object>¶: MetaNames for models creation.

create_version(model, model_uid, meta_props=None)[source]¶

Create a new version for a model.

Note

Model version creation is not allowed for spark models

Parameters

model (str) –
Can be one of following: - path to saved model in format:
- keras (.tgz)
- pmml (.xml)
- scikit-learn (.tar.gz)
- xgboost (.tar.gz)
- tensorflow (.tar.gz)
- spss (.str)
- trained model guid
meta_props (dict/str) –
meta data of the model. To see available meta names use:
```
>>> client.repository.ModelMetaNames.get()
```
model_uid (str) – model ID

Returns

model version details

Return type

dict

Example:

>>> metadata = {
>>>        client.repository.ModelMetaNames.NAME: 'SAMPLE NAME',
>>>        client.repository.ModelMetaNames.DESCRIPTION: 'SAMPLE DESCRIPTION',
>>>        client.repository.ModelMetaNames.AUTHOR_NAME: 'AUTHOR',
>>>        client.repository.ModelMetaNames.FRAMEWORK_NAME: 'pmml',
>>>        client.repository.ModelMetaNames.FRAMEWORK_VERSION: '4.3'
>>>}

Example with local tar.gz containing model:

>>> stored_model_details = client.repository.create_version(path_to_tar_gz, model_uid="MODELID" ,meta_props=metadata)

Example with trained model guid:

>>> stored_model_details = client.repository.create_version(trained_model_guid, model_uid="MODELID" ,meta_props=metadata)

delete(artifact_uid)[source]¶

Delete model, definition, experiment or function from repository.

Parameters: artifact_uid (str) – stored model, definition, experiment or function UID
Returns: returns the status message (“SUCCESS” or FAILED”)
Return type: str

Example:

>>> client.repository.delete(artifact_uid)

delete_definition(artifact_uid)[source]¶

Delete definition from repository.

Parameters: artifact_uid (str) – stored definition UID
Returns: returns the status message (“SUCCESS” or FAILED”)
Return type: str

Example:

>>> client.repository.delete_definition(artifact_uid)

delete_experiment(artifact_uid)[source]¶

Delete experiment definition from repository.

Parameters: artifact_uid (str) – stored experiment UID
Returns: returns the status message (“SUCCESS” or FAILED”)
Return type: str

Example:

>>> client.repository.delete_experiment(artifact_uid)

delete_function(artifact_uid)[source]¶

Delete function from repository.

Parameters: artifact_uid (str) – stored function UID
Returns: returns the status message (“SUCCESS” or FAILED”)
Return type: str

Example:

>>> client.repository.delete_function(artifact_uid)

download(artifact_uid, filename='downloaded_artifact.tar.gz')[source]¶

Download artifact (model or function content) from repository to local file.

Parameters

artifact_uid (str) – stored artifact UID
filename (str) – name of local file to create (optional)

Returns

path to the downloaded file

Return type

str

Note

If filename is not specified, the default filename is “downloaded_artifact.tar.gz”.

Side effect:: save artifact to file.

Example:

>>> client.repository.download(model_uid, 'my_model.tar.gz')

get_definition_details(definition_uid=None, limit=None)[source]¶

Get metadata of stored definitions. If definition uid is not specified returns all model definitions metadata.

Parameters

definition_uid (str) – stored model definition UID (optional)
limit (int) – limit number of fetched records (optional)

Returns

stored definition(s) metadata

Return type

dict (if uid is not None) or {“resources”: [dict]} (if uid is None)

Example:

>>> definition_details = client.repository.get_definition_details(definition_uid)
>>> definition_details = client.repository.get_definition_details()

static get_definition_uid(definition_details)[source]¶

Get uid of stored definition.

Parameters: definition_details (dict) – stored definition details
Returns: uid of stored model
Return type: str

Example:

>>> definition_uid = client.repository.get_definition_uid(definition_details)

static get_definition_url(definition_details)[source]¶

Get url of stored definition.

Parameters: definition_details (dict) – stored definition details
Returns: url of stored definition
Return type: str

Example:

>>> definition_url = client.repository.get_definition_url(definition_details)

get_details(artifact_uid=None)[source]¶

Get metadata of stored artifacts. If uid is not specified returns all models, definitions, experiment, functions, libraries and runtimes metadata.

Parameters: artifact_uid (str) – stored model, definition, experiment, function, library or runtime UID (optional)
Returns: stored artifacts metadata
Return type: dict (if uid is not None) or {“resources”: [dict]} (if uid is None)

Example:

>>> details = client.repository.get_details(artifact_uid)
>>> details = client.repository.get_details()

get_experiment_details(experiment_uid=None, limit=None)[source]¶

Get metadata of stored experiments. If no experiment uid is specified all experiments metadata is returned.

Parameters

experiment_uid (str) – stored experiment UID (optional)
limit (int) – limit number of fetched records (optional)

Returns

stored experiment(s) metadata

Return type

dict (if uid is not None) or {“resources”: [dict]} (if uid is None)

Example:

>>> experiment_details = client.repository.get_experiment_details(experiment_uid)
>>> experiment_details = client.repository.get_experiment_details()

static get_experiment_uid(experiment_details)[source]¶

Get uid of stored experiment.

Parameters: experiment_details (dict) – stored experiment details
Returns: uid of stored experiment
Return type: str

Example:

>>> experiment_uid = client.repository.get_experiment_uid(experiment_details)

static get_experiment_url(experiment_details)[source]¶

Get url of stored experiment.

Parameters: experiment_details (dict) – stored experiment details
Returns: url of stored experiment
Return type: str

Example:

>>> experiment_url = client.repository.get_experiment_url(experiment_details)

get_function_details(function_uid=None, limit=None)[source]¶

Get metadata of function. If no function uid is specified all functions metadata is returned.

Parameters

function_uid (str) – stored function UID (optional)
limit (int) – limit number of fetched records (optional)

Returns

stored function(s) metadata

Return type

dict (if uid is not None) or {“resources”: [dict]} (if uid is None)

Example:

>>> function_details = client.repository.get_function_details(function_uid)
>>> function_details = client.repository.get_function_details()

static get_function_uid(function_details)[source]¶

Get uid of stored function.

Parameters: function_details (dict) – stored function details
Returns: uid of stored function
Return type: str

Example:

>>> function_uid = client.repository.get_function_uid(function_details)

static get_function_url(function_details)[source]¶

Get url of stored function.

Parameters: function_details (dict) – stored function details
Returns: url of stored function
Return type: str

Example:

>>> function_url = client.repository.get_function_url(function_details)

get_model_details(model_uid=None, limit=None)[source]¶

Get metadata of stored models. If model uid is not specified returns all models metadata.

Parameters

model_uid (str) – stored model, definition or pipeline UID (optional)
limit (int) – limit number of fetched records (optional)

Returns

stored model(s) metadata

Return type

dict (if uid is not None) or {“resources”: [dict]} (if uid is None)

Example:

>>> model_details = client.repository.get_model_details(model_uid)
>>> models_details = client.repository.get_model_details()

static get_model_uid(model_details)[source]¶

Get uid of stored model.

Parameters: model_details (dict) – stored model details
Returns: uid of stored model
Return type: str

Example:

>>> model_uid = client.repository.get_model_uid(model_details)

static get_model_url(model_details)[source]¶

Get url of stored model.

Parameters: model_details (dict) – stored model details
Returns: url to stored model
Return type: str

Example:

>>> model_url = client.repository.get_model_url(model_details)

list()[source]¶

List stored models, definitions and experiments. Only first 50 records is shown. For more result use specific list functions.

Returns: None
Return type: None

Note

This function only prints the list of stored models, definitions and experiments

Example:

>>> client.repository.list()

list_definitions(limit=None)[source]¶

List stored definitions. If limit is set to None there will be only first 50 records shown.

Parameters: limit (int) – limit number of fetched records (optional)
Returns: None
Return type: None

Note

This function only prints the list of stored training-definitions

Example:

>>> client.repository.list_definitions()

list_experiments(limit=None)[source]¶

List stored experiments. If limit is set to None there will be only first 50 records shown.

Parameters: limit (int) – limit number of fetched records (optional)
Returns: None
Return type: None

Note

This function only prints the list of stored experiments

Example:

>>> client.repository.list_experiments()

list_functions(limit=None)[source]¶

List stored functions. If limit is set to None there will be only first 50 records shown.

Parameters: limit (int) – limit number of fetched records (optional)
Returns: None
Return type: None

Note

This function only prints the list of stored functions

Example:

>>> client.repository.list_functions()

list_models(limit=None)[source]¶

List stored models. If limit is set to None there will be only first 50 records shown.

Parameters: limit (int) – limit number of fetched records (optional)
Returns: None
Return type: None

Note

This function only prints the list of stored models

Example:

>>> client.repository.list_models()

load(artifact_uid)[source]¶

Load model from repository to object in local environment.

Parameters: artifact_uid (str) – stored model UID
Returns: trained model
Return type: object

Example:

>>> model = client.repository.load(model_uid)

store_definition(training_definition, meta_props)[source]¶

Store training definition into Watson Machine Learning repository on IBM Cloud.

Parameters

training_definition (str) – path to zipped model_definition
meta_props (dict) –
meta data of the training definition. To see available meta names use:
```
>>> client.repository.DefinitionMetaNames.get()
```

Returns

stored training definition details

Return type

dict

Example:

>>> metadata = {
>>>  client.repository.DefinitionMetaNames.NAME: 'my_training_definition',
>>>  client.repository.DefinitionMetaNames.FRAMEWORK_NAME: 'tensorflow',
>>>  client.repository.DefinitionMetaNames.FRAMEWORK_VERSION: '1.5',
>>>  client.repository.DefinitionMetaNames.RUNTIME_NAME: 'python',
>>>  client.repository.DefinitionMetaNames.RUNTIME_VERSION: '3.5',
>>>  client.repository.DefinitionMetaNames.EXECUTION_COMMAND: 'python3 tensorflow_mnist_softmax.py --trainingIters 20'
>>> }
>>> definition_details = client.repository.store_definition(training_definition_filepath, meta_props=metadata)
>>> definition_url = client.repository.get_definition_url(definition_details)

store_experiment(meta_props)[source]¶

Store experiment into Watson Machine Learning repository on IBM Cloud.

param meta_props
meta data of the experiment configuration. To see available meta names use:
>>> client.repository.ExperimentMetaNames.get()
type meta_props

dict

returns

stored experiment details

rtype

dict

Example:

>>> metadata = {
>>>  client.repository.ExperimentMetaNames.NAME: 'my_experiment',
>>>  client.repository.ExperimentMetaNames.EVALUATION_METRICS: ['accuracy'],
>>>  client.repository.ExperimentMetaNames.TRAINING_DATA_REFERENCE: {'connection': {'endpoint_url': 'https://s3-api.us-geo.objectstorage.softlayer.net', 'access_key_id': '***', 'secret_access_key': '***'}, 'source': {'bucket': 'train-data'}, 'type': 's3'},
>>>  client.repository.ExperimentMetaNames.TRAINING_RESULTS_REFERENCE: {'connection': {'endpoint_url': 'https://s3-api.us-geo.objectstorage.softlayer.net', 'access_key_id': '***', 'secret_access_key': '***'}, 'target': {'bucket': 'result-data'}, 'type': 's3'},
>>>  client.repository.ExperimentMetaNames.TRAINING_REFERENCES: [
>>>      {
>>>        'training_definition_url': definition_url_1
>>>      },
>>>      {
>>>        'training_definition_url': definition_url_2
>>>      },
>>>   ],
>>> }
>>> experiment_details = client.repository.store_experiment(meta_props=metadata)
>>> experiment_url = client.repository.get_experiment_url(experiment_details)

store_function(function, meta_props)[source]¶

Store function into Watson Machine Learning repository on Cloud.

As a ‘function’ may be used one of the following:

filepath to gz file
‘score’ function reference, where the function is the function which will be deployed
generator function, which takes no argument or arguments which all have primitive python default values and as result return ‘score’ function

Parameters

meta_props (dict/str) –
meta data or name of the function. To see available meta names use:
```
>>> client.repository.FunctionMetaNames.get()
```
function (str or function) – path to file with archived function content or function (as described above)

Returns

stored function details

Return type

dict

Example: (using score function):

>>> def score(payload):
        values = [[row[0]*row[1]] for row in payload['values']]
        return {'fields': ['multiplication'], 'values': values}
>>> stored_function_details = client.repository.store_function(score, name)

Other, more interesting example is using generator function. In this situation it is possible to pass some variables:

>>> wml_creds = {...}
>>> def gen_function(wml_credentials=wml_creds, x=2):
        def f(payload):
            values = [[row[0]*row[1]*x] for row in payload['values']]
            return {'fields': ['multiplication'], 'values': values}
        return f
>>> stored_function_details = client.repository.store_function(gen_function, name)

In more complicated cases you should create proper metadata, similar to this one:

>>> metadata = {
>>>    client.repository.FunctionMetaNames.NAME: "function",
>>>    client.repository.FunctionMetaNames.DESCRIPTION: "This is ai function",
>>>    client.repository.FunctionMetaNames.RUNTIME_UID: "53dc4cf1-252f-424b-b52d-5cdd9814987f",
>>>    client.repository.FunctionMetaNames.INPUT_DATA_SCHEMA: {"fields": [{"metadata": {}, "type": "string", "name": "GENDER", "nullable": True}]},
>>>    client.repository.FunctionMetaNames.OUTPUT_DATA_SCHEMA: {"fields": [{"metadata": {}, "type": "string", "name": "GENDER", "nullable": True}]},
>>>    client.repository.FunctionMetaNames.TAGS: [{"value": "ProjectA", "description": "Functions created for ProjectA"}]
>>> }
>>> stored_function_details = client.repository.store_function(score, metadata)

store_model(model, meta_props=None, training_data=None, training_target=None, pipeline=None, feature_names=None, label_column_names=None)[source]¶

Store trained model into Watson Machine Learning repository on Cloud.

Parameters

model (object/str) –
Can be one of following:
- The train model object:
  - scikit-learn
  - xgboost
  - spark (PipelineModel)
- path to saved model in format:
  - keras (.tgz)
  - pmml (.xml)
  - scikit-learn (.tar.gz)
  - xgboost (.tar.gz)
  - tensorflow (.tar.gz)
  - spss (.str)
- directory containing model file(s):
  - scikit-learn
  - xgboost
  - tensorflow
- trained model guid
meta_props (dict/str) –
meta data of the training definition. To see available meta names use:
```
>>> client.repository.ModelMetaNames.get()
```
training_data (spark dataframe, pandas dataframe, numpy.ndarray or array) – Spark DataFrame supported for spark models. Pandas dataframe, numpy.ndarray or array supported for scikit-learn models
training_target (array) – array with labels required for scikit-learn models
pipeline (object) – pipeline required for spark mllib models
feature_names (numpy.ndarray or list) – Feature names for the training data in case of Scikit-Learn/XGBoost models. This is applicable only in the case where the training data is not of type - pandas.DataFrame.
label_column_names (numpy.ndarray or list) – Label column names of the trained Scikit-Learn/XGBoost models.

Returns

stored model details

Return type

dict

Note

Keras model content is expected to be an archived vesion of a .h5 file
For deploying a keras model, it is mandatory to pass the FRAMEWORK_LIBRARIES along with other metaprops.
```
>>> client.repository.ModelMetaNames.FRAMEWORK_LIBRARIES: [{'name':'keras', 'version': '2.1.3'}]
```
feature_names is an optional argument containing the feature names for the training data in case of Scikit-Learn/XGBoost models. Valid types are numpy.ndarray and list. This is applicable only in the case where the training data is not of type - pandas.DataFrame.

If the training data is of type pandas.DataFrame and feature_names are provided, feature_names are ignored.

Example:

>>> stored_model_details = client.repository.store_model(model, name)

In more complicated cases you should create proper metadata, similar to this one:

>>> metadata = {
>>>        client.repository.ModelMetaNames.NAME: 'customer satisfaction prediction model',
>>>        client.repository.ModelMetaNames.FRAMEWORK_NAME: 'tensorflow',
>>>        client.repository.ModelMetaNames.FRAMEWORK_VERSION: '1.5',
>>>        client.repository.ModelMetaNames.RUNTIME_NAME: 'python',
>>>        client.repository.ModelMetaNames.RUNTIME_VERSION: '3.5'
>>>}

where FRAMEWORK_NAME may be one of following: “spss-modeler”, “tensorflow”, “xgboost”, “scikit-learn”, “pmml”.

Example with local tar.gz containing model:

>>> stored_model_details = client.repository.store_model(path_to_tar_gz, meta_props=metadata, training_data=None)

Example with local directory containing model file(s):

>>> stored_model_details = client.repository.store_model(path_to_model_directory, meta_props=metadata, training_data=None)

Example with trained model guid:

>>> stored_model_details = client.repository.store_model(trained_model_guid, meta_props=metadata, training_data=None)

update_experiment(experiment_uid, changes)[source]¶

Updates existing experiment metadata.

Parameters

experiment_uid (str) – UID of experiment which definition should be updated
changes (dict) – elements which should be changed, where keys are ExperimentMetaNames

Returns

metadata of updated experiment

Return type

dict

Example

>>> metadata = {
>>> client.repository.ExperimentMetaNames.NAME:"updated_exp"
>>> }
>>> exp_details = client.repository.update_experiment(experiment_uid, changes=metadata)

update_function(function_uid, changes)[source]¶

Updates existing function metadata.

Parameters

function_uid (str) – UID of function which definition should be updated
changes (dict) – elements which should be changed, where keys are FunctionMetaNames

Returns

metadata of updated function

Return type

dict

Example

>>> metadata = {
>>> client.repository.FunctionMetaNames.NAME:"updated_function"
>>> }
>>> function_details = client.repository.update_function(function_uid, changes=metadata)

update_model(model_uid, content_path=None, meta_props=None)[source]¶

Update content of model with new one.

Parameters

model_uid (str) – Model UID
content_path (str) – path to tar.gz with new content of model

Returns

updated metadata of model

Return type

dict

Example

>>> metadata = {
>>> client.repository.ModelMetaNames.NAME:"updated_model"
>>> }
>>> model_details = client.repository.update_model(model_uid, changes=metadata)

class metanames.ModelMetaNames[source]¶

Set of MetaNames for models.

Available MetaNames:

MetaName	Type	Required	Example value
NAME	str	Y	`my_model`
DESCRIPTION	str	N	`my_description`
AUTHOR_NAME	str	N	`John Smith`
FRAMEWORK_NAME	str	N	`tensorflow`
FRAMEWORK_VERSION	str	N	`1.5`
FRAMEWORK_LIBRARIES	list	N	`[{'name': 'keras', 'version': '2.1.3'}]`
RUNTIME_NAME	str	N	`python`
RUNTIME_VERSION	str	N	`3.5`
TRAINING_DATA_SCHEMA	dict	N	`{}`
INPUT_DATA_SCHEMA	dict	N	`{}`
TRAINING_DATA_REFERENCE	dict	N	`{}`
EVALUATION_METHOD	str	N	`multiclass`
EVALUATION_METRICS	list	N	`[{'name': 'accuracy', 'value': 0.64, 'threshold': 0.8}]`
OUTPUT_DATA_SCHEMA	dict	N	{'fields': [{'name': 'ID', 'metadata': {}, 'nullable': True, 'type': 'integer'}, {'name': 'Gender', 'metadata': {}, 'nullable': True, 'type': 'string'}, {'name': 'Status', 'metadata': {}, 'nullable': True, 'type': 'string'}, {'name': 'Children', 'metadata': {}, 'nullable': True, 'type': 'integer'}, {'name': 'Age', 'metadata': {}, 'nullable': True, 'type': 'decimal(14,6)'}, {'name': 'Customer_Status', 'metadata': {}, 'nullable': True, 'type': 'string'}, {'name': 'Car_Owner', 'metadata': {}, 'nullable': True, 'type': 'string'}, {'name': 'Customer_Service', 'metadata': {}, 'nullable': True, 'type': 'string'}, {'name': 'Satisfaction', 'metadata': {}, 'nullable': True, 'type': 'integer'}, {'name': 'Business_Area', 'metadata': {}, 'nullable': True, 'type': 'string'}, {'name': 'prediction', 'metadata': {'modeling_role': 'prediction'}, 'nullable': True, 'type': 'double'}, {'name': 'predictedLabel', 'metadata': {'modeling_role': 'decoded-target', 'values': ['NA', 'Free Upgrade', 'On-demand pickup location', 'Voucher', 'Premium features']}, 'nullable': True, 'type': 'string'}, {'name': 'probability', 'metadata': {'modeling_role': 'probability'}, 'nullable': True, 'type': {'containsNull': True, 'type': 'array', 'elementType': 'double'}}], 'type': 'struct'}
LABEL_FIELD	str	N	`PRODUCT_LINE`
TRANSFORMED_LABEL_FIELD	str	N	`PRODUCT_LINE_IX`
RUNTIME_UID	str	N	`53628d69-ced9-4f43-a8cd-9954344039a8`
TRAINING_DEFINITION_URL	str	N	`https://us-south.ml.cloud.ibm.com/v3/ml_assets/training_definitions/ff24ebec-86af-4f2d-ab4c-2fc9e2a3efbc`

class metanames.ModelDefinitionMetaNames[source]¶

Set of MetaNames for definitions.

Available MetaNames:

MetaName	Type	Required	Example value
NAME	str	Y	`my_training_definition`
DESCRIPTION	str	N	`my_description`
TRAINING_DATA_REFERENCES	list	N	`{}`
AUTHOR_NAME	str	N	`John Smith`
FRAMEWORK_NAME	str	Y	`tensorflow`
FRAMEWORK_VERSION	str	Y	`1.5`
RUNTIME_NAME	str	Y	`python`
RUNTIME_VERSION	str	Y	`3.5`
EXECUTION_COMMAND	str	N	`python3 tensorflow_mnist_softmax.py --trainingIters 20`

class metanames.ExperimentMetaNames[source]¶

Set of MetaNames for experiments.

Available MetaNames:

MetaName	Type	Required	Example value
NAME	str	Y	`Hand-written Digit Recognitionu`
DESCRIPTION	str	N	`Hand-written Digit Recognition training`
TAGS	list	N	`[{'value': 'dsx-project.<project-guid>', 'description': 'DSX project guid'}]`
AUTHOR_NAME	str	N	`John Smith`
EVALUATION_METHOD	str	N	`multiclass`
EVALUATION_METRICS	list	N	`['accuracy']`
TRAINING_REFERENCES	list	Y	`[{'training_definition_url': 'https://ibm-watson-ml.mybluemix.net/v3/ml_assets/training_definitions/12345', 'compute_configuration': {'name': 'k80'}}, {'training_definition_url': 'https://ibm-watson-ml.mybluemix.net/v3/ml_assets/training_definitions/67890'}]`
TRAINING_DATA_REFERENCE	dict	Y	`{'connection': {'endpoint_url': 'https://s3-api.us-geo.objectstorage.softlayer.net', 'access_key_id': '*', 'secret_access_key': '*'}, 'source': {'bucket': 'train-data'}, 'type': 's3'}`
TRAINING_RESULTS_REFERENCE	dict	Y	`{'connection': {'endpoint_url': 'https://s3-api.us-geo.objectstorage.softlayer.net', 'access_key_id': '*', 'secret_access_key': '*'}, 'target': {'bucket': 'result-data'}, 'type': 's3'}`

class metanames.FunctionMetaNames[source]¶

Set of MetaNames for AI functions.

Available MetaNames:

MetaName	Type	Required	Example value
NAME	str	Y	`ai_function`
DESCRIPTION	str	N	`This is ai function`
RUNTIME_UID	str	N	`53628d69-ced9-4f43-a8cd-9954344039a8`
INPUT_DATA_SCHEMA	dict	N	`{'type': 'struct', 'fields': [{'name': 'x', 'type': 'double', 'nullable': False, 'metadata': {}}, {'name': 'y', 'type': 'double', 'nullable': False, 'metadata': {}}]}`
OUTPUT_DATA_SCHEMA	dict	N	`{'type': 'struct', 'fields': [{'name': 'multiplication', 'type': 'double', 'nullable': False, 'metadata': {}}]}`
TAGS	list	N	`[{'value': 'ProjectA', 'description': 'Functions created for ProjectA'}]`

deployments ¶

class client.Deployments(client)[source]¶

Deploy and score published artifacts (models and functions).

create(artifact_uid=None, name='Artifact deployment', description='Description of deployment', asynchronous=False, deployment_type='online', deployment_format='Core ML', meta_props=None, **kwargs)[source]¶

Create deployment (online) from artifact. As artifact we understand model or function which may be deployed.

Parameters

artifact_uid (str) – Published artifact UID (model or function UID)
name (str) – Deployment name
description (str) – Deployment description
asynchronous (bool) – if False then will wait until deployment will be fully created before returning
deployment_type (str) – type of deployment (‘online’, ‘virtual’). Default one is ‘online’
deployment_format (str) – file format of virtual deployment. Currently supported is ‘Core ML’ only (default value)
meta_props (dict) – dictionary with parameters used for virtual deployment (Core ML format)

Returns

details of created deployment

Return type

dict

Example

>>> online_deployment = client.deployments.create(model_uid, 'Deployment X', 'Online deployment of XYZ model.')
>>> virtual_deployment = client.deployments.create(model_uid, 'Deployment A', 'Virtual deployment of XYZ model.', deployment_type='virtual')

delete(deployment_uid)[source]¶

Delete deployment.

Parameters: deployment_uid (str) – Deployment UID
Returns: returns the status message (“SUCCESS” or FAILED”)
Return type: str

Example:

>>> client.deployments.delete(deployment_uid)

download(virtual_deployment_uid, filename=None)[source]¶

Downloads file deployment of specified UID. Currently supported format is Core ML.

Parameters

virtual_deployment_uid (str) – UID of virtual deployment
filename (str) – filename of downloaded archive (optional)

Returns

path to downloaded file

Return type

str

get_details(deployment_uid=None, limit=None)[source]¶

Get information about your deployment(s).

Parameters

deployment_uid (str) – Deployment UID (optional)
limit (int) – limit number of fetched records (optional)

Returns

metadata of deployment(s)

Return type

dict (if uid is not None) or {“resources”: [dict]} (if uid is None)

Example

>>> deployment_details = client.deployments.get_details(deployment_uid)
>>> deployment_details = client.deployments.get_details(deployment_uid=deployment_uid)
>>> deployments_details = client.deployments.get_details()

static get_download_url(deployment_details)[source]¶

Get deployment_download_url from deployment details.

Parameters: deployment_details (dict) – Created deployment details
Returns: deployment download URL that is used to get file deployment (for example: Core ML)
Return type: str

Example

>>> deployment_url = client.deployments.get_download_url(deployment)

static get_scoring_url(deployment)[source]¶

Get scoring_url from deployment details.

Parameters: deployment (dict) – Created deployment details
Returns: scoring endpoint URL that is used for making scoring requests
Return type: str

Example

>>> scoring_url = client.deployments.get_scoring_url(deployment)

get_status(deployment_uid)[source]¶

Get status of deployment creation.

Parameters: deployment_uid (str) – Guid of deployment
Returns: status of deployment creation
Return type: str

Example

>>> status = client.deployments.get_status(deployment_uid)

static get_uid(deployment_details)[source]¶

Get deployment_uid from deployment details.

Parameters: deployment_details (dict) – Created deployment details
Returns: deployment UID that is used to manage the deployment
Return type: str

Example

>>> deployment_uid = client.deployments.get_uid(deployment)

get_uids()[source]¶

Get all deployments uids.

Returns: list of uids
Return type: list of strings

Example:

>>> deployments_uids = client.deployments.get_uids()

static get_url(deployment_details)[source]¶

Get deployment_url from deployment details.

Parameters: deployment_details (dict) – Created deployment details
Returns: deployment URL that is used to manage the deployment
Return type: str

Example

>>> deployment_url = client.deployments.get_url(deployment)

list(limit=None)[source]¶

List deployments. If limit is set to None there will be only first 50 records shown.

Parameters: limit (int) – limit number of fetched records (optional)
Returns: None
Return type: None

Note

This function only prints the list of deployments

Example:

>>> client.deployments.list()

score(scoring_url, payload, transaction_id=None)[source]¶

Make scoring requests against deployed artifact.

Parameters

scoring_url (str) – scoring endpoint URL
payload (dict) – records to score
transaction_id (str) – transaction id to be passed with records during payload logging (optional)

Returns

scoring result containing prediction and probability

Return type

dict

Example:

>>> scoring_payload = {'fields': ['GENDER','AGE','MARITAL_STATUS','PROFESSION'], 'values': [['M',23,'Single','Student'],['M',55,'Single','Executive']]}
>>> predictions = client.deployments.score(scoring_url, scoring_payload)

update(deployment_uid, name=None, description=None, asynchronous=False, meta_props=None)[source]¶

Update model used in deployment to the latest version. The scoring_url remains. Name and description change will not work for online deployment. For virtual deployments the file will be updated under the same download_url.

Parameters

deployment_uid (str) – Deployment UID
name (str) – new name for deployment
description (str) – new description for deployment
meta_props (dict) – dictionary with parameters used for virtual deployment (Core ML format)

Returns

updated metadata of deployment

Return type

dict

Example

>>> metadata = {
>>> client.deployments.ConfigurationMetaNames.NAME:"updated_Deployment"
>>> }
>>> deployment_details = client.deployments.update(deployment_uid, changes=metadata)

training ¶

class client.Training(client)[source]¶

Train new models.

cancel(run_uid)[source]¶

Cancel model training.

Parameters: run_uid (str) – ID of trained model

Example

>>> client.training.cancel(run_uid)

delete(run_uid)[source]¶

Delete training run.

Parameters: run_uid (str) – ID of trained model
Returns: returns the status message (“SUCCESS” or FAILED”)
Return type: str

Example

>>> client.training.delete(run_uid)

get_definition_details(definition_uid=None, limit=None)[source]¶

Get metadata of stored definitions. If definition uid is not specified returns all model definitions metadata.

Parameters

definition_uid (str) – stored model definition UID (optional)
limit (int) – limit number of fetched records (optional)

Returns

stored definition(s) metadata

Return type

dict

Example

>>> definition_details = client.training.get_definition_details(definition_uid)
>>> definition_details = client.training.get_definition_details()

get_definition_uid(definition_details)[source]¶

Get uid of stored definition.

Parameters: definition_details (dict) – stored definition details
Returns: uid of stored model
Return type: str

Example

>>> definition_uid = client.training.get_definition_uid(definition_details)

get_definition_url(definition_details)[source]¶

Get url of stored definition.

Parameters: definition_details (dict) – stored definition details
Returns: url of stored definition
Return type: str

Example

>>> definition_url = client.training.get_definition_url(definition_details)

get_details(run_uid=None, limit=None)[source]¶

Get training run details.

Parameters

run_uid (str) – ID of training run (optional, if not provided all runs details are returned)
limit (int) – limit number of fetched records (optional)

Returns

training run(s) details

Return type

dict (if uid is not None) or {“resources”: [dict]} (if uid is None)

Example

>>> training_run_details = client.training.get_details(run_uid)
>>> training_runs_details = client.training.get_details()

get_frameworks()[source]¶

Get list of supported frameworks.

Returns: supported frameworks for training
Return type: dict

Example

>>> model_details = client.training.get_frameworks()

get_latest_metrics(run_uid)[source]¶

Get latest metrics values.

Parameters: run_uid ({0}) – ID of trained model
Returns: metric values
Return type: list of dicts

Example

>>> client.training.get_latest_metrics(run_uid)

get_metrics(run_uid)[source]¶

Get metrics values.

Parameters: run_uid (str) – ID of trained model
Returns: metric values
Return type: list of dicts

Example

>>> client.training.get_metrics(run_uid)

static get_run_uid(run_details)[source]¶

Get uid of training run.

Parameters: run_details (dict) – training run details
Returns: uid of training run
Return type: str

Example

>>> model_uid = client.training.get_run_uid(run_details)

static get_run_url(run_details)[source]¶

Get training run url from training run details.

Parameters: run_details (dict) – Created training run details
Returns: training run URL that is used to manage the training
Return type: str

Example

>>> run_url = client.training.get_run_url(run_details)

get_status(run_uid)[source]¶

Get training status.

Parameters: run_uid (str) – ID of trained model
Returns: training run status
Return type: dict

Example

>>> training_status = client.training.get_status(run_uid)

list(limit=None)[source]¶

List training runs. If limit is set to None there will be only first 50 records shown.

Returns: None
Return type: None

Note

This function only prints the list of training runs

Example

>>> client.training.list()

list_definitions(limit=None)[source]¶

List stored definitions. If limit is set to None there will be only first 50 records shown.

Returns: None
Return type: None

Note

This function only prints the list of stored training-definitions

Example:

>>> client.training.list_definitions()

monitor_logs(run_uid)[source]¶

Monitor training log file (prints log content to console).

Parameters: run_uid (str) – ID of trained model

Example

>>> client.training.monitor_logs(run_uid)

monitor_metrics(run_uid)[source]¶

Monitor metrics log file (prints log content to console).

Parameters: run_uid (str) – ID of trained model

Example

>>> client.training.monitor_metrics(run_uid)

run(definition_uid, meta_props, asynchronous=True)[source]¶

Train new model.

Parameters

definition_uid (str) – uid to saved model_definition/pipeline
meta_props (dict) –
meta data of the training configuration. To see available meta names use:
```
>>> client.training.ConfigurationMetaNames.show()
```
asynchronous (bool) – Default True means that training job is submitted and progress can be checked later. False - method will wait till job completion and print training stats.

Returns

training run details

Return type

dict

Example

>>> metadata = {
>>>  client.training.ConfigurationMetaNames.NAME: u'Hand-written Digit Recognition',
>>>  client.training.ConfigurationMetaNames.TRAINING_DATA_REFERENCE: {
>>>          u'connection': {
>>>              u'endpoint_url': u'https://s3-api.us-geo.objectstorage.service.networklayer.com',
>>>              u'access_key_id': u'***',
>>>              u'secret_access_key': u'***'
>>>          },
>>>          u'source': {
>>>              u'bucket': u'wml-dev',
>>>          }
>>>          u'type': u's3'
>>>      }
>>> client.training.ConfigurationMetaNames.TRAINING_RESULTS_REFERENCE: {
>>>          u'connection': {
>>>              u'endpoint_url': u'https://s3-api.us-geo.objectstorage.service.networklayer.com',
>>>              u'access_key_id': u'***',
>>>              u'secret_access_key': u'***'
>>>          },
>>>          u'target': {
>>>              u'bucket': u'wml-dev-results',
>>>          }
>>>          u'type': u's3'
>>>      },
>>> client.training.ConfigurationMetaNames.COMPUTE_CONFIGURATION: {'name': 'p100'},
>>> client.training.ConfigurationMetaNames.EXECUTION_COMMAND: "python3 tensorflow_mnist_softmax.py --trainingIters 20",
>>> }
>>> run_details = client.training.run(definition_uid, meta_props=metadata)
>>> run_uid = client.training.get_run_uid(run_details)

class metanames.TrainingConfigurationMetaNames[source]¶

Set of MetaNames for trainings.

Available MetaNames:

MetaName	Type	Required	Default value	Example value
NAME	str	Y		`Hand-written Digit Recognition`
DESCRIPTION	str	N		`Hand-written Digit Recognition training`
AUTHOR_NAME	str	N		`John Smith`
TRAINING_DATA_REFERENCE	dict	Y		`{'connection': {'endpoint_url': 'https://s3-api.us-geo.objectstorage.softlayer.net', 'access_key_id': '*', 'secret_access_key': '*'}, 'source': {'bucket': 'train-data'}, 'type': 's3'}`
TRAINING_RESULTS_REFERENCE	dict	Y		`{'connection': {'endpoint_url': 'https://s3-api.us-geo.objectstorage.softlayer.net', 'access_key_id': '*', 'secret_access_key': '*'}, 'target': {'bucket': 'train-data'}, 'type': 's3'}`
EXECUTION_COMMAND	str	N	`<value from model definition>`	`python3 tensorflow_mnist_softmax.py --trainingIters 20`
COMPUTE_CONFIGURATION	dict	N	`{'name': 'k80'}`	`{'name': 'k80'}`

experiments ¶

class client.Experiments(client)[source]¶

Run new experiment.

ConfigurationMetaNames = <watson_machine_learning_client.metanames.ExperimentMetaNames object>¶: MetaNames for experiments creation.

delete(experiment_run_uid)[source]¶

Delete experiment run.

Parameters: experiment_run_uid (str) – experiment run UID
Returns: returns the status message (“SUCCESS” or FAILED”)
Return type: str

Example:

>>> client.experiments.delete(experiment_run_uid)

Hint

This function is only for deleting experiment runs. To delete whole experiment use:

>>> client.repository.delete(experiment_uid)

get_definition_details(experiment_uid=None, limit=None)[source]¶

Get metadata of stored experiments. If neither experiment uid nor url is specified all experiments metadata is returned.

Parameters

experiment_uid (str) – stored experiment UID (optional)
limit (int) – limit number of fetched records (optional)

Returns

stored experiment(s) metadata

Return type

dict

Example

>>> experiment_details = client.experiments.get_definition_details(experiment_uid)
>>> experiment_details = client.experiments.get_definition_details()

get_definition_uid(experiment_details)[source]¶

Get uid of stored experiment.

Parameters: experiment_details (dict) – stored experiment details
Returns: uid of stored experiment
Return type: str

Example

>>> experiment_uid = client.experiments.get_definition_uid(experiment_details)

get_definition_url(experiment_details)[source]¶

Get url of stored experiment.

Parameters: experiment_details (dict) – stored experiment details
Returns: url of stored experiment
Return type: str

Example

>>> experiment_url = client.experiments.get_definition_url(experiment_details)

get_details(experiment_uid=None, limit=None)[source]¶

Get metadata of experiment(s) run(s). If no experiment_uid is provided, runs will be listed for all existing experiments.

Parameters

experiment_uid (str) – experiment UID (optional)
limit – limit number of fetched records (optional) :type limit: int

Returns

experiment(s) run(s) metadata

Return type

dict (if uid is not None) or {“resources”: [dict]} (if uid is None)

Example:

>>> experiment_details = client.experiments.get_details(experiment_uid)
>>> experiment_details = client.experiments.get_details()

get_latest_metrics(experiment_run_uid)[source]¶

Get latest metrics values for experiment run.

Parameters: experiment_run_uid (str) – ID of training run
Returns: metric values
Return type: list of dicts

Example

>>> client.experiments.get_latest_metrics(experiment_run_uid)

get_metrics(experiment_run_uid)[source]¶

Get all metrics values for experiment run.

Parameters: experiment_run_uid (str) – ID of training run
Returns: metric values
Return type: list of dicts

Example

>>> client.experiments.get_metrics(experiment_run_uid)

get_run_details(experiment_run_uid)[source]¶

Get metadata of particular experiment run.

Parameters: experiment_run_uid (bool) – experiment run UID
Returns: experiment run metadata
Return type: dict

Example:

>>> experiment_run_details = client.experiments.get_run_details(experiment_run_uid)

static get_run_uid(experiment_run_details)[source]¶

Get experiment run uid.

Parameters: experiment_run_details (object) – details of experiment run
Returns: experiment run uid
Return type: str

Example:

>>> experiment_run_uid = client.experiments.get_run_uid(experiment_run_details)

static get_run_url(experiment_run_details)[source]¶

Get experiment run url.

Parameters: experiment_run_details (object) – details of experiment run
Returns: experiment run url
Return type: str

Example:

>>> experiment_run_url = client.experiments.get_run_url(experiment_run_details)

get_status(experiment_run_uid)[source]¶

Get experiment status.

Parameters: experiment_run_uid (bool) – ID of experiment run
Returns: experiment status
Return type: dict

Example:

>>> experiment_status = client.experiments.get_status(experiment_run_uid)

static get_training_runs(experiment_run_details)[source]¶

Get experiment training runs details.

Parameters: experiment_run_details – details of experiment run
Type: object
Returns: training runs
Return type: array

Example:

>>> training_runs = client.experiments.get_training_runs(experiment_run_details)

static get_training_uids(experiment_run_details)[source]¶

Get experiment training uids.

Parameters: experiment_run_details (object) – details of experiment run
Returns: training uids
Return type: array

Example:

>>> training_uids = client.experiments.get_training_uids(experiment_run_details)

list_definitions(limit=None)[source]¶

List stored experiments. If limit is set to None there will be only first 50 records shown.

Returns: None
Return type: None

Note

This function only prints the list of experiments

Example:

>>> client.experiments.list_definitions()

list_runs(experiment_uid=None, limit=None)[source]¶

List experiment runs. If experiment_uid set to None and limit set to None only 50 first records will be displayed.

Parameters

experiment_uid (str) – experiment UID (optional)
limit (int) – limit number of fetched records (optional)

Returns

None

Return type

None

Note

This function only prints the list of experiments-runs

Example:

>>> client.experiments.list_runs()
>>> client.experiments.list_runs(experiment_uid)

list_training_runs(experiment_run_uid)[source]¶

List training runs triggered by experiment run.

Parameters: experiment_run_uid (str) – experiment run UID
Returns: None
Return type: None

Note

This function only prints the list of training-runs associated with an experiment-run.

Example:

>>> client.experiments.list_training_runs(experiment_run_uid)

monitor_logs(experiment_run_uid)[source]¶

Monitor experiment run log files (prints log content to console).

Parameters: experiment_run_uid (str) – ID of experiment run

Example:

>>> client.experiments.monitor_logs(experiment_run_uid)

monitor_metrics(experiment_run_uid)[source]¶

Monitor metrics log file (prints metrics to console).

Parameters: experiment_run_uid – ID of experiment run

Example

>>> client.experiments.monitor_metrics(experiment_run_uid)

run(experiment_uid, asynchronous=True)[source]¶

Run experiment.

Parameters

experiment_uid (str) – ID of stored experiment
asynchronous (bool) – Default True means that experiment is started and progress can be checked later. False - method will wait till experiment end and print experiment stats.

Returns

experiment run details

Return type

dict

Example:

>>> experiment_run_status = client.experiments.run(experiment_uid)
>>> experiment_run_status = client.experiments.run(experiment_uid, asynchronous=False)

learning_system ¶

class client.LearningSystem(client)[source]¶

Continuous Learning System.

get_details(model_uid)[source]¶

Get details of learning system.

Parameters: model_uid (str) – ID of model for this learning system
Returns: learning system details
Return type: dict

Example:

>>> learning_system_details = client.learning_system.get_details(model_uid)

get_metrics(model_uid)[source]¶

Get evaluation metrics.

Parameters: model_uid (str) – ID of stored model

Example:

>>> runs_details = client.learning_system.get_metrics(model_uid)

get_run_details(run_uid)[source]¶

Get run details.

Parameters: run_uid (str) – ID of learning iteration run

Example:

>>> run_details = client.learning_system.get_run_details(run_uid)

get_run_href(run_details)[source]¶

Get href of run (learning iteration).

Parameters: run_details (dict) – details of run
Returns: run href
Return type: str

Example:

>>> run_uid = client.learning_system.get_run_href(run_details)

get_run_uid(run_details)[source]¶

Get uid of run (learning iteration).

Parameters: run_details (dict) – details of run
Returns: run uid
Return type: str

Example:

>>> run_uid = client.learning_system.get_run_uid(run_details)

get_runs(model_uid)[source]¶

Get learning iterations (runs).

Parameters: model_uid (str) – ID of stored model
Returns: Learning Iterations
Return type: dict

Example:

>>> runs_details = client.learning_system.get_runs(model_uid)

list()[source]¶

List existing learning systems.

Returns: None
Return type: None

Note

This function only prints the list of existing learning_systems

Example:

>>> client.learning_system.list()

list_metrics(model_uid)[source]¶

Get evaluation metrics.

param model_uid

ID of stored model

type model_uid

str

Returns: None
Return type: None

Note

This function only prints the evaluation metrics

Example:

>>> runs_details = client.learning_system.get_metrics(model_uid)

list_runs(model_uid=None)[source]¶

List learning iterations.

Parameters: model_uid (str) – ID of stored model (optional)
Returns: None
Return type: None

Note

This function only prints the list of learning iterations

Example:

>>> client.learning_system.list_runs()

run(model_uid, asynchronous=True)[source]¶

Run learning iterations.

Parameters

model_uid (str) – ID of stored model
asynchronous (bool) – if the run should be asynchronous (optional)

Return type

dict

Example:

>>> client.learning_system.run(model_uid)
>>> client.learning_system.run(model_uid, asynchronous=False)

send_feedback(model_uid, feedback_data, fields=None)[source]¶

Send feedback data to learning system.

Parameters

model_uid (str) – ID of model
feedback_data (list of rows) – rows of feedback data to be send
fields (list of strings) – list of fields (optional)

Example:

>>> client.learning_system.send_feedback(model_uid, [["a1", 1, 1.0], ["a2", 2, 3.4]])
>>> client.learning_system.send_feedback(model_uid, [["a1", 1.0], ["a2", 3.4]], fields=["id", "value"])

setup(model_uid, meta_props)[source]¶

Setup continuous learning system for stored model.

Parameters

model_uid ({str_type}) – ID of stored model
meta_props (dict) – learning system configuration meta data

Returns

learning system configuration details

Return type

dict

Example:

>>> learning_system = client.learning_system.create(model_uid, meta_props=configuration)

update(model_uid, changes)[source]¶

Updates existing learning system metadata.

Parameters

model_uid (str) – UID of model which learning system should be updated
changes (dict) – elements which should be changed, where keys are LearningSystemMetaNames

Returns

metadata of updated learning system

Return type

dict

Example

>>> metadata = {
>>> client.learning_system.ConfigurationMetaNames.NAME:"updated_function"
>>> }
>>> learning_system_details = client.learning_system.update(model_uid, changes=metadata)

class metanames.LearningSystemMetaNames[source]¶

Set of MetaNames for learning system.

Available MetaNames:

MetaName	Type	Required	Example value
FEEDBACK_DATA_REFERENCE	dict	Y	`{}`
SPARK_REFERENCE	dict	N	`{}`
MIN_FEEDBACK_DATA_SIZE	int	Y	`100`
AUTO_RETRAIN	str	Y	`conditionally`
AUTO_REDEPLOY	str	Y	`always`
COMPUTE_CONFIGURATION	dict	N	`{'name': 'k80'}`
TRAINING_RESULTS_REFERENCE	dict	N	`{'connection': {'endpoint_url': 'https://s3-api.us-geo.objectstorage.softlayer.net', 'access_key_id': '*', 'secret_access_key': '*'}, 'target': {'bucket': 'train-data'}, 'type': 's3'}`

runtimes ¶

class client.Runtimes(client)[source]¶

Creates Runtime Specs and associated Custom Libraries.

delete(runtime_uid, with_libraries=False)[source]¶

Delete runtime.

Parameters

runtime_uid (str) – Runtime UID
autoremove (bool) – if set to False, only runtime will be removed, if set to True, all libraries belonging only to it will be removed

Returns

returns the status message (“SUCCESS” or FAILED”)

Return type

str

Example:

>>> client.runtimes.delete(runtime_uid)
>>> client.runtimes.delete(runtime_uid, with_libraries=True)

delete_library(library_uid)[source]¶

Delete custom library from repository.

Parameters: library_uid (str) – stored custom library UID
Returns: returns the status message (“SUCCESS” or FAILED”)
Return type: str

Example:

>>> client.runtimes.delete_library(library_uid)

download_configuration(runtime_uid, filename=None)[source]¶

Downloads configuration file for runtime with specified UID.

Parameters

runtime_uid (str) – UID of runtime
filename (str) – filename of downloaded archive (optional)

Returns

path to the downloaded file

Return type

str

Note

If filename is not specified, the default filename is “runtime_configuration.yaml”.

Side effect:: save runtime configuration to file.

Example:

>>> filename = client.runtimes.download_configuration(runtime_uid)
>>> client.runtimes.download_configuration(runtime_uid, filename=filename)

download_library(library_uid, filename=None)[source]¶

Downloads custom library content with specified UID.

Parameters

library_uid (str) – UID of library
filename (str) – filename of downloaded archive (optional)

Returns

path to the downloaded file

Return type

str

Note

If filename is not specified, the default filename is “<LIBRARY-NAME>-<LIBRARY-VERSION>.zip”.

Side effect:: save library content to file.

Example:

>>> filename = client.runtimes.download_library(library_uid)
>>> client.runtimes.download_library(library_uid, filename=filename)

get_details(runtime_uid=None, limit=None)[source]¶

Get information about your runtime(s).

Parameters

runtime_uid (str) – Runtime UID (optional)
limit (int) – limit number of fetched records (optional)

Returns

metadata of runtime(s)

Return type

dict (if uid is not None) or {“resources”: [dict]} (if uid is None)

Example:

>>> runtime_details = client.runtimes.get_details(runtime_uid)
>>> runtime_details = client.runtimes.get_details(runtime_uid=runtime_uid)
>>> runtime_details = client.runtimes.get_details()

get_library_details(library_uid=None, limit=None)[source]¶

Get information about your custom libraries(s).

Parameters

library_uid (str) – Library UID (optional)
limit (int) – limit number of fetched records (optional)

Returns

metadata of libraries(s)

Return type

dict (if uid is not None) or {“resources”: [dict]} (if uid is None)

Example:

>>> library_details = client.runtimes.get_library_details(library_uid)
>>> library_details = client.runtimes.get_library_details(library_uid=library_uid)
>>> library_details = client.runtimes.get_library_details()

static get_library_uid(library_details)[source]¶

Get library uid from library details.

Parameters: library_details (dict) – Created library details
Returns: library uid
Return type: str

Example:

>>> library_uid = client.runtimes.get_library_uid(library_details)

static get_library_url(library_details)[source]¶

Get library url from library details.

Parameters: library_details (dict) – Created library details
Returns: library url
Return type: str

Example:

>>> library_url = client.runtimes.get_library_url(library_details)

static get_uid(details)[source]¶

Get runtime uid from runtime details, or runtime uid from model details.

Parameters: details (dict) – Created runtime or model details
Returns: runtime uid
Return type: str

Example:

>>> runtime_uid = client.runtimes.get_uid(runtime_details)

static get_url(details)[source]¶

Get runtime url from runtime details, or runtime url from model details.

Parameters: details (dict) – Created runtime or model details
Returns: runtime url
Return type: str

Example:

>>> runtime_url = client.runtimes.get_url(runtime_details)

list(limit=None)[source]¶

List runtimes. If limit is set to None there will be only first 50 records shown.

Parameters: limit (int) – limit number of fetched records (optional)
Returns: None
Return type: None

Note

This function only prints the list of stored runtimes

Example:

>>> client.runtimes.list()

list_libraries(runtime_uid=None, limit=None)[source]¶

List custom libraries. If limit is set to None there will be only first 50 records shown.

Parameters

runtime_uid (str) – Runtime UID
limit (int) – limit number of fetched records (optional)

Returns

None

Return type

None

Note

This function only prints the list of stored libraries

Example:

>>> client.runtimes.list_libraries()
>>> client.runtimes.list_libraries(runtime_uid)

store(meta_props)[source]¶

Create runtime.

Parameters

meta_props (dict) –

dictionary with parameters describing runtime spec. To see available meta names use:

>>> client.runtimes.ConfigurationMetaNames.get()

Returns

details of created runtime

Return type

dict

The simplest way you might use me is:

>>> runtime_details = client.runtimes.store({
    client.runtimes.ConfigurationMetaNames.NAME: "test",
    client.runtimes.ConfigurationMetaNames.PLATFORM: {"name": "python", "version": "3.5"}
})

The most complex way to use me is:

>>>
    # here library is created
    lib_meta = {
        client.runtimes.LibraryMetaNames.NAME: "libraries_custom",
        client.runtimes.LibraryMetaNames.DESCRIPTION: "custom libraries for scoring",
        client.runtimes.LibraryMetaNames.FILEPATH: "/home/user/my_lib.zip",
        client.runtimes.LibraryMetaNames.VERSION: "1.0",
        client.runtimes.LibraryMetaNames.PLATFORM: {"name": "python", "versions": ["3.5"]}
    }
    custom_library_details = client.runtimes.store_library(lib_meta)
    custom_library_uid = client.runtimes.get_library_uid(custom_library_details)
    # now, metaprops for runtime spec are prepared
    meta = {
        client.runtimes.ConfigurationMetaNames.NAME: "runtime_spec_python_3.5",
        client.runtimes.ConfigurationMetaNames.DESCRIPTION: "test",
        client.runtimes.ConfigurationMetaNames.PLATFORM: {
            "name": "python",
            "version": "3.5"
        },
        client.runtimes.ConfigurationMetaNames.LIBRARIES_DEFINITIONS: [ # here in the background are created additional two libraries
            LibraryDefinition("my_lib_1", "1.0", "/home/user/my_lib_1.zip", description="t", platform={"name": "python", "versions": ["3.5"]}),
            LibraryDefinition("my_lib_2", "1.1", "/home/user/my_lib_2.zip")
        ],
        client.runtimes.ConfigurationMetaNames.LIBRARIES_UIDS: [custom_library_uid] # already existing lib is linked here
    }
    # Now runtime spec is created. Note that during runtime spec creation also "my_lib_1" and "my_lib_2" will be created and "libraries_custom" will be linked.
    runtime_details = client.runtimes.store(meta)

store_library(meta_props)[source]¶

Create custom library.

Parameters

meta_props (dict) –

dictionary with parameters describing custom library. To see available meta names use:

>>> client.runtimes.LibraryMetaNames.get()

Returns

details of created custom library

Return type

dict

Example:

>>> library_details = client.runtimes.store_library({
           client.runtimes.LibraryMetaNames.NAME: "libraries_custom",
           client.runtimes.LibraryMetaNames.DESCRIPTION: "custom libraries for scoring",
           client.runtimes.LibraryMetaNames.FILEPATH: custom_library_path,
           client.runtimes.LibraryMetaNames.VERSION: "1.0",
           client.runtimes.LibraryMetaNames.PLATFORM: {"name": "python", "versions": ["3.5"]}
       })

class metanames.RuntimeMetaNames[source]¶

Set of MetaNames for Runtime Specs.

Available MetaNames:

MetaName	Type	Required	Example value
NAME	str	Y	`runtime_spec_python_3.5`
DESCRIPTION	str	N	`py35`
PLATFORM	dict	Y	`{'name': 'python', 'version': '3.5'}`
LIBRARIES_UIDS	list	N	`['46dc9cf1-252f-424b-b52d-5cdd9814987f']`
LIBRARIES_DEFINITIONS	list	N	`[{'name': 'libraries_custom', 'description': 'custom libraries for scoring', 'filepath': '/home/lib.gz', 'version': '1.0', 'platform': 'python', 'platform_versions': ['3.5']}]`
CONFIGURATION_FILEPATH	str	N	`/home/env_config.yaml`

class metanames.LibraryMetaNames[source]¶

Set of MetaNames for Custom Libraries.

Available MetaNames:

MetaName	Type	Required	Example value
NAME	str	Y	`my_lib`
DESCRIPTION	str	N	`my lib`
PLATFORM	dict	Y	`{'name': 'python', 'versions': ['3.5']}`
VERSION	str	Y	`1.0`
FILEPATH	str	Y	`/home/user/my_lib_1_0.zip`

Welcome to watson-machine-learning-client’s documentation!¶

Installation ¶

Requirements ¶

Supported machine learning frameworks ¶

Sample notebooks ¶

API ¶

service_instance ¶

repository ¶

deployments ¶

training ¶

experiments ¶

learning_system ¶

runtimes ¶

Indices and tables ¶

Navigation

Related Topics