IBM Watson™ Discovery Service Ideas

We've moved...

You'll be redirected shortly, we've moved to our new idea portal: https://ibm-watson.ideas.aha.io

Make it possible to explicitly kick off the ranker training process

Currently the ranker model building kicks off some indeterminate amount of time after training samples are added to WDS.  Would like the ability to:

  1. add a batch of training data
  2. trigger the training process
  3. poll the collection details api endpoint until model has finished building
  4. run a set of queries and collect result relevancy metrics
  5. add more training data
  6. poll until training is done
  7. re-run test queries and collect result relevancy metrics
  8. ... repeat until all the training batches are added...
  9. make plots of (improved) relevancy performance on the test set as training set size is increased

Right now -- after step 5, I poll the api and have to wait a really really long time.  There are two separate delays
(1) delay while waiting for the model training process to begin (no idea what schedules this, it seems like maybe its time based)

(2) delay while waiting for training to complete.  

 

The second delay is usually much less than the first for my datasets.  If i could trigger the training process I could eliminate the first delay entirely.

  • Nathaniel Cohen
  • Jan 9 2018
  • Attach files