Hi
I am new in HTM and trying to use NUPIC as a part of my thesis implementation. So I need some help, how can I find the most appropriate encoder? In my thesis data streams in network have to be analysed to detect anomaly behaviours.
Thanks in advance.
Which encoder(s) depends on the data type(s) of your streams. This doc on encoders has a lot of info:
http://nupic.docs.numenta.org/1.0.3/api/algorithms/encoders.html#
The HTM school episodes on encoders might be a good place to start too if you haven’t seen them already:
I think the main types of encoders commonly used are for scalars, categories, datetimes and coordinates in space. There are demos using each I believe, though most commonly using a scalar stream and a datetime stream, combined using the multi-encoder. If your stream(s) could be boiled down into some collection of these types that’d be the quickest way to get it running, though its also possible to design a specialty encoder from your datatype, as long as you can define what trait(s) would make input values more and less similar to each other.
3 Likes
Thank you dear Sam.
I 've seen HTM school episode about encoders, it was useful and gave me basic information about different types of encoders, but I prefer to design my own. Is there any source to show how I can start to do it step by step?
1 Like
Here’s the thread you should check out, just made a day or so ago
1 Like