HTM Studio (Awesome App!)

cogmission · June 25, 2019, 4:24pm

Hi,

Congratulations on HTM Studio! I’m really impressed with it! I found it fun and easy to use - kudos!

I got really excited when I saw something mentioned in your blog about a “Param Finder”, and I have a few questions:

Is HTM Studio open source?
Can the Param Finder source be shared? Could it be ported to HTM.Java?
To what degree is it or could it be a substitute for swarming?

Anyway, thanks for making such a cool HTM product. I can’t wait to promote it and use it as an example of things you can do with NuPIC!

Cheers,
David

rhyolight · September 30, 2016, 4:07pm

Yes, you can see it here: https://github.com/numenta/numenta-apps/tree/master/unicorn

@cmaver I think we should update http://numenta.com/htm-studio so that it says something about HTM Studio being open source and link to the repo.

It is here I think, and sure you can port it.

It is not a replacement for swarming, because it has a narrow use case. See here for detailed info:

github.com

numenta/numenta-apps/blob/master/unicorn/py/unicorn_backend/param_finder.py#L184-L213


"""
Find parameters for a given time series dataset with heuristics.


@param samples Sequence of two tuples (timestamp, value), where
  timestamp of type datetime.datetime and value is a number (int of float)


@return: JSON object with the following properties:


  "aggInfo" aggregation information, JSON null if no aggregation is needed
    otherwise, a JSON object contains the folowing properties:
      "windowSize": aggregation window size in seconds (integer)
      "func": A string representation of the aggregation function (string)
        the following values are supported: "sum", "mean"


  "modelInfo": JSON object describing the model configuration that
  contains the following properties:
    "modelConfig": OPF Model Configuration parameters (JSON object) for
    passing to the OPF "ModelFactory.create()" method as the
    "modelConfig" parameter.

This file has been truncated. show original

Basically, it only works to find optimal aggregation and model params for one scalar stream of data (with a timestamp). I believe @ycui wrote this code, so he might have more to say.

cmaver · September 30, 2016, 5:06pm

Thanks @rhyolight and @cogmission.
Good suggestion. I will update the page to reference it being open source.

marion · September 30, 2016, 6:40pm

Thanks for the feedback @cogmission @rhyolight’s answer is great. As he mentioned, it’s only a replacement for swarming if you have a single stream of scalar data. I will only add that if you are interested in porting the param finder, then you might want to look into: https://github.com/numenta/numenta-apps/blob/master/unicorn/tests/py/integration/compatibility_test/model_runner_2_compatibility_test.py

It’s a compatibility test to ensure that we get comparable results on NAB data when using the NAB’s params or the params generated by the param finder. I think it’s a useful test to port as well, if you are thinking of porting param_finder.py to HTM java.

cogmission · September 30, 2016, 9:15pm

You’re welcome. Thanks for pointing out the existence of that compatibility test. I believe that will come in handy if/when it becomes obvious that I’m going to port the param finder. For now my brain is NAB-fried… with butter on top! So I need a rest…

Anyway, great job on HTM Studio!

Matheus_Araujo · October 19, 2018, 12:12pm

Cogmission, did you really port the Param Finder?
In case you didn’t do it yet , are you interested in doing it?
Could I help you in that case?