What’s new in Decision Optimization for Cloud Pak for Data 2.5
IBM Cloud Pak for Data 2.5 has just been released. This post describes what’s new and what’s been improved for Decision Optimization.
IBM Cloud Pak™ for Data (CP4D) is a fully-integrated data and AI platform that modernizes how businesses collect, organize and analyze data and infuse AI throughout their organizations.
Among all the options that are available is IBM Watson Studio Premium, a ready-to-use enterprise set of tools to accelerate productivity for data science and machine learning workflows for training and deployment. This includes Decision Optimization (DO).
The Decision Optimization experience
Decision Optimization provides the same user experience in IBM Cloud Pak for Data as in Watson Studio and Watson Machine Learning Cloud. This means that you can:
- develop optimization models in Python using Jupyter notebooks,
- develop, debug and tune optimization models using a dedicated model builder, with visualizations and scenarios,
- develop optimization models starting from your data and with natural language using a modeling assistant,
- deploy models and integrate their execution in production applications.
Watson Machine Learning deployment
Just as with Watson Studio Cloud, model deployment is now based on Watson Machine Learning (WML). The mechanism and APIs are similar to what is described in this post.
Models can be deployed and used in production using either a Python API or a REST API.
The main differences for CP4D are credentials and deployment space configurations.
Credentials do not use IBM Cloud API keys, but usernames and passwords. For example, you can use the Python API as follows:
CPD_cluster = 'MY_URL'
username = "MY_LOGIN"
password = "MY_PASSWORD"
wml_credentials = {
"username": username,
"password": password,
"instance_id" : "wml_local",
"url": CPD_cluster,
"version": "2.5.0"
}client = WatsonMachineLearningAPIClient(wml_credentials)
Also, deployment for CP4D is not based on instances but on deployment spaces, hence the following additional call:
# Find the space ID
space_name = 'MY_SPACE'
space_id = [x['metadata']['id'] for x in client.spaces.get_details()['resources'] if x['entity']['name'] == space_name][0]
client.set.default_space(space_id)Different deployment spaces can be created for different projects or different sets of deployements. User should be granted access control to the corresponding deployment space to be able to deploy and use models from this space.
Deployment User Interface
Models can also be deployed using a new User Interface.
From the model builder, you can directly save a model for deployment from the scenario you have validated and want to deploy.
Then, the saved model is available in the project and can be promoted to a deployment space.
You can also open the saved model to see information such as the input and output schemas.
After the model is promoted to a deployment space, new deployments can be created, configuring the number of nodes and hardware definition. Refer to this post for an explanation of these configurations.
Monitoring User Interface
All jobs created and executed on a model deployment can be monitored from the User Interface.
Other enhancements
While the main change is the deployment of models using WML, many other improvements have been introduced for Decision Optimization.
The model builder design has been improved for better productivity. Steps have been reduced to Prepare Data, Run Model, and Explore Solution.
Dashboards have been renamed to Visualizations, which can be used to validate and compare scenarios.
The support of OPL models has been improved, and the model builder now supports multiple files models, with enhanced Python and OPL editors.
Finally, it is possible to work with connected data just as described here for Watson Studio Cloud.
