On Sat, Apr 30, 2016 at 4:53 PM, Ryan J. McCall <[email protected]> wrote: > 1. How much support does HTM engine provide for swarming? Can a set of model > params be sent to the engine with each model creation request? I found this > issue https://github.com/numenta/numenta-apps/issues/503 but it doesn't link > to a PR -- was this feature implemented?
None, and no. I re-opened https://github.com/numenta/numenta-apps/issues/503, it was not implemented. > 2. For the HTM for Stocks app was swarming performed for each metric? (seems > time consuming but doable) Or was a generic set of params (e.g., NAB anomaly > params) used and the metric trained on an initial batch of data? No, HTM Engine uses only one set of model params (with minor tweaks for min/max) for all models. Since HTM for Stocks uses HTM Engine, it just uses the models provided for it. They are a generic set of model params tuned for anomaly detection of single-field scalar input. > 3. It seems only anomaly scores are managed by HTM engine and not > predictions. It seems like there was some interest in this > (http://lists.numenta.org/pipermail/nupic_lists.numenta.org/2015-December/012440.html) > but it wasn't implemented. Would you accept a PR for this? (should it go to > the skeleton-htmengine-app?) I would love a PR against https://github.com/numenta/numenta-apps/issues/504, which I just re-opened. I closed the tickets mentioned above because it didn't seem that anyone was interested in the functionality or actually doing the work to implement it. If you worked on them and submitted PRs, I'll try to get them reviewed. Regards, Matt
