Frameworks and software specifications in Watson Machine Learning

Framework

A framework contains a set of reusable algorithms, tools, and libraries for developing machine learning models that you can build upon for specific application requirements. Frameworks describe the machine learning or deep learning framework that is used to build the model. You can use popular frameworks such as Tensorflow, which supports both deep learning and traditional machine learning algorithms with Watson Machine Learning for more efficient development.

Model type

Model type indicates the machine learning or deep learning framework and the framework version that is used to build the machine learning model. Machine learning or deep learning models that you build by using the same framework might not be compatible across versions. For instance, if you built your model with pytorch-onnx_1.10 model type, you must build your model by using pytorch version 1.10 and save the model in ONNX format.

Software specification

Software specification define the programming language and version that you use for building a model or a function. You can use software specifications to configure the software that is used for running your models and functions. You can also define the software version to be used and include your own extensions. For example, you can use conda .yml files or custom libraries.

Platform

Platform describes the hardware platform on which you can deploy your machine learning model with a supported software specification. Depending on the framework with wich you built your model, you can use Watson Machine Learning to deploy your model on Z (s390x), Power (PPC), or x86 platforms.

Lifecycle of supported frameworks and software specifications

Frameworks and software specifications that are supported when you install the latest version of Cloud Pak for Data as a Service go through deprecation and discontinuation phases in future releases of Cloud Pak for Data as a Service, described as follows:

1. Supported: The runtime environment corresponding to the software specification is supported when you install or upgrade to the latest version of Cloud Pak for Data.

   For example, the software specification runtime-24.1-py3.11 is supported in Cloud Pak for Data version 5.0 In this case, if you install or upgrade to a new Watson Machine Learning service instance, you can create or deploy machine learning models that use runtime-24.1-py3.11.

   For more information, see Supported deployment frameworks and software specifications.

2. Deprecated: The runtime environment corresponding to the software specification is supported when you install the latest version of Cloud Pak for Data as a Service, but will be removed in an upcoming release.

   For example, the software specification runtime-22.2-py3.10 is deprectaed in Cloud Pak for Data as a Service. In this case, if you install a new Watson Machine Learning service instance, you can create or deploy machine learning models that use runtime-22.2-py3.10. However, the support for runtime-22.2-py3.10 will be discontinued in an upcoming release.

3. Discontinued: The runtime environment corresponding to the software specification is not supported when you install or upgrade to the latest version of Cloud Pak for Data as a Service.

   For example, the support for software specification based on Python 3.9 is discontinued for Cloud Pak for Data as a Service. In this case, if you install a new Watson Machine Learning service instance, you cannot create or deploy machine learning models that use software specification based on Python 3.9.

Managing outdated frameworks and software specifications

Update software specifications and frameworks in your models when they become outdated. Sometimes, you can seamlessly update your assets. In other cases, you must retrain or redeploy your assets.

For more information, see Managing outdated software specifications or frameworks.

Parent topic: Deploying assets with Watson Machine Learning