Predictive Intelligence - New classification solution available to all with Vancouver release?

AmenaA · ‎06-24-2024

For point 1 -> New classification solution available to all - Where do we expect to see them?

Thanks in Advance

Brian Bakker · ‎07-01-2024

There are 4 new system properties starting with [glide.platform_ml.api.enable_workflow_], one for each ML Capability (Classification, Similarity, Clustering and Regression) to switch from the old ML Java Trainer to the new ML Python Trainer, which has the new and improved Classifier engine, and you will no longer need to associate a Word Corpus with the solution definition, as it will be using GUSE (Google Universal Sentence Encoder) in our data centers for all Word Corpora -

To use the new Classifier engine, the system property [glide.platform_ml.api.enable_workflow_classification] should be set to "true". If the system property does not exist in Vancouver, it can be manually added. When upgrading to Washington, all 4 system properties will exist, and be set to "true" so that it will automatically use the new ML Python Trainer.

Hope this helps.

Regards,

Brian

View solution in original post

Joe Walters · 3 weeks ago

Does this mean that we don't need to configure Word Corpus at all, or just that we don't need to associate them within our Solutions? (I noticed the Word Corpus field is no longer available on Solution forms, but there's still a left nav module for Word Corpus.)

Lener Pacania1 · 3 weeks ago

@Joe Walters - correct word corpus no longer needed. See docs.

Brian Bakker · 3 weeks ago

@Joe Walters

You no longer need to configure a Word Corpus for any PI solution when using the Workflow capabilities. The Word Corpus module is still there in case you need to revert back to non-workflow capabilities, by setting these system properties to "false", which will also add the Word Corpus field back onto the solution definition form.

Reason is that it simplifies the creation of your solution definitions, and the in-house Word Corpus that uses GUSE has shown to increase correct predictions by up to 10% compared to using a Word Corpus defined by our customers on the solution definition when using the non-workflow capability.

I hope that answers your question.

Regards,

Brian