Skip to content

Latest commit

 

History

History
93 lines (67 loc) · 3.46 KB

02-build-bento-service.md

File metadata and controls

93 lines (67 loc) · 3.46 KB
Raw

7.2 Building Your Prediction Service with BentoML

Links

Notes

We need BentoML model store to save an XGBoost model instance in the following way:

# Save xgboost model
bentoml.xgboost.save_model('credit_risk_model',
                            model,
                            custom_objects={'DictVectorizer': dv})

BentoML will generator a directory in the home directory where the model will be kept.

Once the model is saved, we can create a service.py file that will be used to define the BentoML service:

import bentoml
from bentoml.io import JSON


# Pull the model as model reference (it pulls all the associate metadata of the model)
model_ref = bentoml.xgboost.get('credit_risk_model:latest')
# Call DictVectorizer object using model reference
dv = model_ref.custom_objects['DictVectorizer']
# Create the model runner (it can also scale the model separately)
model_runner = model_ref.to_runner()

# Create the service 'credit_risk_classifier' and pass the model
svc = bentoml.Service('credit_risk_classifier', runners=[model_runner])


# Define an endpoint on the BentoML service
@svc.api(input=JSON(), output=JSON()) # decorate endpoint as in json format for input and output
def classify(application_data):
    # transform data from client using dictvectorizer
    vector = dv.transform(application_data)
    # make predictions using 'runner.predict.run(input)' instead of 'model.predict'
    prediction = model_runner.predict.run(vector)
    
    result = prediction[0] # extract prediction from 1D array
    print('Prediction:', result)

    if result > 0.5:
        return {'Status': 'DECLINED'}
    elif result > 0.3:
        return {'Status': 'MAYBE'}
    else:
        return {'Status': 'APPROVED'}

Once the service and the endpoint is created we can run the app using the command: bentoml serve service:svc, where service is the script and svc is the service name.

There are some key things to consider at the time of bentoml version 1.0.7:

  • The final model has to be trained without providing feature names in the DMatrix constructor to prevent "ValueError" while building the BentoML application.
  • When running bento service instead of using http://0.0.0.0:3000/, use http://localhost:3000/.
  • On Windows systems bentoml serve service:svc --reload is not working. The issue has been added here.

Add notes from the video (PRs are welcome)

⚠️ The notes are written by the community.
If you see an error here, please create a PR with a fix.

Additional Issue Support

  • Thanks for watching! Depending on your local setup, we do find issues from time to time. If you run into anything strange we have a big community of BentoML users who would be happy to receive issue feedback: BentoML slack community. And if you're around shoot me a direct message and say hi! 😃

~Tim

Navigation