Hi Wakan, We have a Python file which is setup to run each algorithm by piecing them together and manually handing data from the output of one to the input of the next. It processes days of the week and you can see it learn and start to predict the days of the week. It may require some tweaks to get working as it was written last year and I'm not sure if it will collide with any recent NuPIC changes... but it is really good for experimenting because it has all the algorithms and shows how their connected and how they function (ScalarEncoder, SP, TM, CLAClassifier - no Anomaly though...) :-P
Check it out: https://gist.github.com/cogmission/101323115e70bb6671d3 Cheers, David On Fri, May 6, 2016 at 6:42 AM, Wakan Tanka <[email protected]> wrote: > Hello David, > thank you this was useful, in fact my understanding was very similar to > what you wrote. One more question: can I somehow experiment directly with > TemporalMemory? I'm using OPF but if it's not possible via OPF is there > something else? Or even better: can I directly experiments with OPF's > components? My understanding of OPF is that it covers many aspects of > anomaly detection (data encoding, HTM, CLA, metrics ...) what is the best > way to learn about each of those processes by getting your hands dirty? > Thank you. > > On Fri, May 6, 2016 at 12:15 PM, cogmission (David Ray) < > [email protected]> wrote: > >> Hi Wakan, >> >> I believe reset and resetSequenceStates do the same thing (i.e. they're >> eventually passed in to the TemporalMemory). Reset is for the >> TemporalMemory directly; and resetSequenceStates is for the model as you >> say. >> >> They're both used to indicate the beginning of a new Sequence. When >> transitioning between days of the week for instance, A week beginning >> Sunday and ending Saturday, **could** have reset called before cycling >> again to Sunday, but not necessarily. If reset is called, the TM will not >> learn the transition from Saturday (end of week), to Sunday (beginning of >> next week) - other wise the transition would be learned. >> >> Hope that helps a bit... >> >> >> >> On Fri, May 6, 2016 at 5:05 AM, Wakan Tanka <[email protected]> wrote: >> >>> Hello NuPIC. >>> 1. Is reset flag [1] somehow related to resetSequenceStates [2] method? >>> I assume reset flag [1] is used during swarm and resetSequenceStates [2] >>> during model run (because resetSequenceStates is called on model). Is this >>> correct? >>> 3. What exactly is resetSequenceStates [2] good for? Documentation [2] >>> says: "Signal that the input record is the start of a new sequence." I >>> guess that all previously learned patterns are NOT forgotten (otherwise the >>> resetSequenceStates [2] will not have purpose and new model instance would >>> be sufficient) it just signalizes the new start of sequence (the same >>> sequence which was learn). IMHO one use cas is e.g. in experiment (that >>> I've seen somewhere but I cannot find it right now :( ) where audio was >>> pushed to NuPIC and then NuPIC was queried to play what he remembered. Then >>> the same audio was pushed to NuPIC again and NuPIC was once again queried >>> to play what he remembered and so on and so on. After each push the output >>> from NuPIC was more like than original audio because. So audio was pushed >>> to NuPIC and then I guess reset was called to signalize that the same >>> sequence will start again (what would happen if reset was not called). Is >>> this correct? >>> 4. What exactly is reset flag [1] good for? I cannot imagine why I would >>> want to use it. Is there any use case scenario? >>> >>> [1] https://github.com/numenta/nupic/wiki/NuPIC-Input-Data-File-Format >>> [2] >>> http://numenta.org/docs/nupic/classnupic_1_1frameworks_1_1opf_1_1model_1_1_model.html#a7ecda56dff89a402d9e21730bd5f7226 >>> >>> -- >>> Best Regards >>> >>> Name: Wakan Tanka a.k.a. Wakatana a.k.a. MackoP00h >>> Location: Europe >>> Note: I'm non native English speaker so please bare with me ;) >>> Contact: >>> [email protected] >>> http://stackoverflow.com/users/1616488/wakan-tanka >>> https://github.com/wakatana >>> https://twitter.com/MackoP00h >>> >>> >> >> >> -- >> *With kind regards,* >> >> David Ray >> Java Solutions Architect >> >> *Cortical.io <http://cortical.io/>* >> Sponsor of: HTM.java <https://github.com/numenta/htm.java> >> >> [email protected] >> http://cortical.io >> > > > > -- > Best Regards > > Name: Wakan Tanka a.k.a. Wakatana a.k.a. MackoP00h > Location: Europe > Note: I'm non native English speaker so please bare with me ;) > Contact: > [email protected] > http://stackoverflow.com/users/1616488/wakan-tanka > https://github.com/wakatana > https://twitter.com/MackoP00h > -- *With kind regards,* David Ray Java Solutions Architect *Cortical.io <http://cortical.io/>* Sponsor of: HTM.java <https://github.com/numenta/htm.java> [email protected] http://cortical.io
