Align C++ and Py implementation params for all classes

[breznak]

Params in both implementations must have same

• name
• default value
• same meaning :)
• ideally same order of params in functions
• should be of same type where possible
• functions return same type

TODO: check

• [ ] TP
• [ ] TM
• [ ] SP
• [ ] Classifier
• [ ] same SP init params
• [ ] Encoders
• [ ] Anomaly, ...

[david-ragazzi]

:+1:

[breznak]

I'd like to fix the first parts: that all classes (in both implementations) use optimal params. @mrcslws @subutai please advice how to obtain such parameters? (ideally future-proof for further modifications)

[subutai]

I don't know that there is a single set of parameters that are optimal for everything. For global inhibition and typical streaming analytics scenarios, the parameters here are an excellent starting point:

$nupic/src/nupic/frameworks/opf/common_models/anomaly_params_random_encoder/best_single_metric_anomaly_params.json

[breznak]

Thank you for the file, that's the one I've been looking for. Definitely not a single set of best params overall, but best of what we can do for a generic domain.

[breznak]

src/nupic/frameworks/opf/common_models/anomaly_params_random_encoder/best_single_metric_anomaly_params.json

Btw, do you have any experience how those params are (over)fitted? Will it work OK with ScalarEncoder, or RDSE with slightly different params, etc..?

[andrewmalta13]

@breznak, you can check off TM. https://github.com/numenta/nupic/pull/3254

[breznak]

great! Thanks for headsup @andrewmalta13 :clap: