Anamoly detection with HTM


I am planning to develop an anomaly detection model with HTM.
Most of the examples given in the repo are performing anomaly detection where the input is a single column of numerical quantities.
But In my case, I am thinking of developing an anomaly detection model where each row will be a combination of numerical and categorical variables. I sort of want to combine them into a single encoding and do anomaly detection at a record level.
Are there any examples similar to this particular use case, or any articles I can look at to get started on this front.


1 Like

Here’s a thread where I give a somewhat detailed overview on implementing multiple fields. I haven’t actually implemented the category encoder myself, so it’d be great if anyone could point to an example of a params file that has it. I know that the scalar encoder can be configured to act as a category encoder, by converting each category to a number and setting ‘n’ and ‘w’ such that there’s no overlap between different inputs.

Luckily, @lscheinkman is working on a new category encoder example.

1 Like