H2O-3 Extensions for Integration with H2O.ai Cloud

A python library containing API extensions for interaction with various cloud components. The library currently supports integration with:

• MLOps

Install

Install prerequisites to your Python environment (Jupyter kernel):

pip install h2o
pip install https://s3.amazonaws.com/artifacts.h2o.ai/releases/ai/h2o/mlops/rel-0.58.0/5/h2o_mlops_client-0.58.0%2B42d986c.rel0.58.0.5-py2.py3-none-any.whl

Install the library:

# TODO UPDATE url:
pip install https://h2o-release.s3.amazonaws.com/h2o/cloud-extensions/h2o_cloud_extensions-3.39.0.99999-py2.py3-none-any.whl

Configure

Open a Python environment and import h2o with cloud extensions and create or connect to h2o cluster

import h2o
import h2o_cloud_extensions as hce
h2o.init()

Set a H2O Cloud instance you will be connecting to:

hce.settings.connection.client_id = 'hac-platform-public'
hce.settings.connection.token_endpoint_url = 'https://auth.internal.dedicated.h2o.ai/auth/realms/hac/protocol/openid-connect/token'

Get authenticated against H2O.ai cloud (https://internal.dedicated.h2o.ai/auth/get-platform-token) and set platform token:

hce.settings.connection.refresh_token = "TOKEN_THAT_YOU_RECIEVED_AFTER_AUTHENTICATION"

MLOps

Set MLOps instance and the project that you will utilize for publishing your models.

hce.settings.mlops.api_url = 'https://mlops-api.internal.dedicated.h2o.ai'
hce.settings.mlops.project_name = 'My-project-for-h2o-3-models'

If the project with a given name does not exist, it will be automatically created. In that case, it would be convenient to set also a project description.

hce.settings.mlops.project_description = 'H2O-3 is so cool. This will be fun!'

MLOps Integration

H2O-3 cloud extension library is able to publish and deploy MOJO models trained via a single H2O algorithm (estimator) Grid Search and AutoML.

Model Trained via a Single Algorithm (Estimator)

Train a model:

from h2o.estimators import H2OGradientBoostingEstimator

prostate = h2o.import_file("http://s3.amazonaws.com/h2o-public-test-data/smalldata/prostate/prostate.csv")
prostate["CAPSULE"] = prostate["CAPSULE"].asfactor()
predictors = ["ID","AGE","RACE","DPROS","DCAPS","PSA","VOL","GLEASON"]
response = "CAPSULE"

pros_gbm = H2OGradientBoostingEstimator(nfolds=5,
                                        seed=1111,
                                        keep_cross_validation_predictions = True)
                                        
pros_gbm.train(x=predictors, y=response, training_frame=prostate)

Model Publishing

To manually publish a trained model, call:

pros_gbm.publish()

The above method call could be performed automatically straight after the model is trained if the below is set:

hce.settings.mlops.estimator.automatic_publishing = True

To check whether the model was published or not, call:

pros_gbm.is_published()

Model Deployment

To manually deploy a trained model, call:

pros_gbm.deploy(environment = "DEV")

The environment does not have to be specified if the below is set:

hce.settings.mlops.deployment_environment = "DEV"

A model could be automatically deployed to an environment specified with the above settings straight after publishing the model. To achieve that, set the below:

hce.settings.mlops.estimator.automatic_deployment = True

To check whether the model was deployed to a given environment, call:

pros_gbm.is_deployed(environment = "DEV")

Model Trained via Grid Search

Train models:

from h2o.estimators import H2OGradientBoostingEstimator
from h2o.grid.grid_search import H2OGridSearch

prostate = h2o.import_file("http://s3.amazonaws.com/h2o-public-test-data/smalldata/prostate/prostate.csv")
prostate["CAPSULE"] = prostate["CAPSULE"].asfactor()
predictors = ["ID","AGE","RACE","DPROS","DCAPS","PSA","VOL","GLEASON"]
response = "CAPSULE"
gbm_params = {'learn_rate': [i * 0.01 for i in range(1, 11)],
                'max_depth': list(range(2, 11)),
                'sample_rate': [i * 0.1 for i in range(5, 11)],
                'col_sample_rate': [i * 0.1 for i in range(1, 11)]}

search_criteria = {'strategy': 'RandomDiscrete', 'max_models': 5, 'seed': 2}

gbm_grid = H2OGridSearch(model=H2OGradientBoostingEstimator,
                          hyper_params=gbm_params,
                          search_criteria=search_criteria)
                          
gbm_grid.train(x=predictors, y=response, training_frame=prostate)

Model Publishing

To manually publish all trained models, call:

gbm_grid.publish()

The above method call could be performed automatically straight after all models are trained if the below is set:

hce.settings.mlops.grid_search.automatic_publishing = True

To check whether all models were published, call:

gbm_grid.is_published()

Model Deployment

To manually deploy all published models, call:

gbm_grid.deploy(environment = "DEV")

The environment does not have to be specified if you specify the below:

hce.settings.mlops.deployment_environment = "DEV"

Models could be automatically deployed to an environment specified with the above settings straight after publishing all models. To achieve that, set the below:

hce.settings.mlops.grid_search.automatic_deployment = True

To check whether all models were deployed to a given environment, call:

gbm_grid.is_deployed(environment = "DEV")

Model Trained via AutoML

Run AutoML to train models:

from h2o.automl import H2OAutoML

train = h2o.import_file("https://s3.amazonaws.com/h2o-public-test-data/smalldata/higgs/higgs_train_10k.csv")
x = train.columns
y = "response"
x.remove(y)
train[y] = train[y].asfactor()

aml = H2OAutoML(max_models=5, seed=1)

aml.train(x=x, y=y, training_frame=train)

Model Publishing

To manually publish all trained models, call:

aml.publish(strategy="all")

The parameter stragegy is set to best by default, which publishes only the best performing model from leaderboard. The other option is to set the parameter to "all", which publishes all leaderboard models. The default parameter value could be changed by the below command:

hce.settings.mlops.automl.publishing_strategy = "all"

The call of aml.publish() could be performed automatically straight after model training is finished if the below is set:

hce.settings.mlops.automl.automatic_publishing = True

To check whether all models were published (Change to stragegy="best" if you're interested only in the best model), call:

aml.is_published(strategy="all")

Model Deployment

To manually deploy all published models, call:

aml.deploy(environment = "DEV")

The environment does not have to be specified if you specify the below:

hce.settings.mlops.deployment_environment = "DEV"

Models could be automatically deployed to an environment specified with the above settings straight after publishing all models. To achieve that, set the below:

hce.settings.mlops.grid_search.automatic_deployment = True

To check whether all models published models were deployed to a given environment, call:

aml.is_deployed(environment = "DEV")