Deploying Natural Language Processing models in Watson Machine Learning
You can deploy a Natural Language Processing model by using Python functions or Python scripts. Both online and batch deployments are supported.
- The prediction output returned from
<model>.run()
is an object of a class specific to the concerned data model's prediction class (for example,watson_nlp.data_model.syntax.SyntaxPrediction
). Such objects cannot be serialized into JSON format so the prediction output must be converted to either the Python dictionary type or JSON by using the<prediction output>.to_dict()
(recommended) or<prediction output>.to_json()
methods. If you don't convert the output, scoring API requests will return an error. - If the
Watson NLP
add-on is installed, you can access the location of pre-trained models in the Python function code by using theLOAD_PATH
environment variable. - Prediction input payload and prediction response returned from
score()
must meet the requirements listed in online scoring and jobs API documentation. - Scoring requests for NLP models may fail with an
Out of Memory
error reported by the underlying JVM runtime. If this happens, patch the deployment to use a hardware specification with more available memory.
Supported Platforms:
List of platforms supporting the deployment of NLP models:
- x86-64
Supported Software Specifications
List of software specifications supporting the deployment of NLP models:
runtime-22.1-py3.9
- Custom software specifications based on
runtime-22.1-py3.9
runtime-22.2-py3.10
- Custom software specifications based on
runtime-22.2-py3.10
NLP model deployment examples
For examples, refer to these Jupyter notebooks:
- Car complaint analysis
- Complaint classification
- Deploying a pretrained sentiment model
- Entity extraction on financial complaints
- Financial complaint analysis
Parent topic: Deploying assets