Skip to main content
Version: v1.0.x

Model registry

This example demonstrates how to use the eScorer Python client to interact with the model registry. You can download the complete example here.

Prerequisites

Install the H2O eScorer Python client.

List all models​

The model.list method fetches the list of models stored in the model registry. Subsequent calls to this method will return the model list from the cache.

models = await client.model.list()
[Model(name='pipeline-telco.mojo', version='1.10.4.1', id='be1c5024-a170-11ed-a236-aa09dd2aa20a'),
 Model(name='pipelineSF.mojo', version='1.8.4.1', id='87ab50b8-6a04-11ea-a272-0242ac110002'),
 Model(name='gbm.zip', version='3.20.0.5', id='-949234640338551712'),
 Model(name='riskmodel.mojo', version='1.9.0', id='0a4bbd12-dcad-11ea-ab05-0242ac110002'),
 Model(name='credit_card_pipeline.mojo', version='1.10.4.1', id='a96fe76a-9916-11ed-8835-36e42284ede4')]

To force a refresh of the model list, set the force_refresh parameter to True.

models = await client.model.list(force_refresh=True)

If you want the client to only fetch model details when required, set the lazy_load parameter to True.

models = await client.model.list(lazy_load=True)

Get model details​

The model_registry.get method fetches the details of a specific model from the model registry.

model = await client.model_registry.get('riskmodel.mojo')
model
Model(name='riskmodel.mojo', version='1.9.0', id='0a4bbd12-dcad-11ea-ab05-0242ac110002')

In addition to the id, name, and version, the model object also contains the features attributes which provides the list of features and their data types used by the model.

model.features
[Feature(name='loan_amnt', type='Float32'),
 Feature(name='term', type='Str'),
 Feature(name='int_rate', type='Float32'),
 Feature(name='installment', type='Float32'),
 Feature(name='emp_length', type='Float32'),
 Feature(name='annual_inc', type='Float32'),
 Feature(name='verification_status', type='Str'),
 Feature(name='addr_state', type='Str'),
 Feature(name='dti', type='Float32'),
 Feature(name='inq_last_6mths', type='Float32'),
 Feature(name='revol_bal', type='Float32'),
 Feature(name='revol_util', type='Float32'),
 Feature(name='total_acc', type='Float32')]

Upload a model​

The model_registry.upload method uploads a model to the model registry and returns the model object.

model = await client.model_registry.upload(
    model_path="/path/to/riskmodel.mojo",
    model_name="riskmodel.mojo",
    validate=True,
    replace=True,
)

model_name is optional and defaults to the model file name.

If validate is set to True, the client will validate the model before uploading it.

If replace is set to True, the client will replace the existing model with the same name.

Rename a model​

The model_registry.rename method renames a model in the model registry.

model = await client.model_registry.rename(
    model_name="riskmodel.mojo",
    new_model_name="riskmodel_v1.mojo",
)

Delete a model​

The model_registry.delete method deletes a model from the model registry.

await client.model_registry.delete("riskmodel.mojo")
Feedback