ML Similarity Search

With great interest I’ve been reading the blog on similarity search.
https://www.stardog.com/blog/similarity-search/

I have a couple of questions

  • Will null variables work, iow can optionals be used in the select ?
  • If I would like to use inferencing, e.g. classes which are subclasses of other learned statements, should I use the inferencing to create multiple entries in the trainingdata ?
  • how do updates work, e.g. learning with new data?

Bart

Hi Bart,

Null variables and optionals
Null variables are supported, and are interpreted as the training instance having no value for the feature. This is naturally interpreted by the similarity metric, and the recommended way to model unknown features.
You can use optionals when selecting data for input into the model. When doing a predict query, just make sure that you use a sub-query. We have some examples of more complex queries in this github repo: stardog-examples.

Inference
Inference results can be given as features to the model. Since inference can generate many results for a single variable, make sure that you use sets to aggregate all the data into a single result.

Model Updates
Right now we don’t support incremental learning, which means that if the underlying data changes, you have to rebuild the model. This can be done by overwriting or deleting and creating a new model.

-pedro

1 Like

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.