NUP-2487: Update category prediction example

numenta / nupic-legacy

Numenta Platform for Intelligent Computing is an implementation of Hierarchical Temporal Memory (HTM), a theory of intelligence based strictly on the neuroscience of the neocortex.

http://numenta.org/

GNU Affero General Public License v3.0

6.34k stars 1.56k forks source link

NUP-2487: Update category prediction example #3799

Closed lscheinkman closed 6 years ago

lscheinkman commented 6 years ago

Fixes #3790 Replaces the Word Prediction example with a simple Web site link prediction example based on MSNBC.com Anonymous Web Data Data Set

The new example demonstrate how to use the CategoryEncoder and HTMPrediction model to predict the next page the user is more likely to click based on the page category.

@rhyolight Please review CC: @subutai

rhyolight commented 6 years ago

looking

subutai commented 6 years ago

it does not use the SP, which seems to make it an atypical use case

With category inputs, the SP does not add anything. It will actually make the results a little worse.

subutai commented 6 years ago

So I think no-SP should be typical for category examples, as long as there are no scalar inputs. Maybe we should explain that?

rhyolight commented 6 years ago

I would be ok with this example if we:

exposed the category encoding more clearly (use strings, not ints, print the "string" ==> 0000011111111000000000000 somewhere during the run so users can see how their strings are being encoded
explained why the SP is not necessary
removed the artificial looping over users during training (I always tell people we don't need training data, and this makes it seem like we do)
explained how we came up with the model parameters and any category-specific parameters settings

lscheinkman commented 6 years ago

@rhyolight Ready for review

lscheinkman commented 6 years ago

@rhyolight One more review