Anomaly Detection Using ML.Net

Swapnil Dave 41

I want to detect anomaly for the time series i am having and i want to understand how does the microsoft detects anomaly withe the different algorithms it provides.

The problem or the challenge i am facing is that the method arguments. I don't know what arguments should i pass in the algorithm.

Like For Example:

DetectAnomalyBySrCnn has arguments like :- trainingWindowSize, threshold, backAddWindowSize, and many more.

How to pass these parameters in a way that i get correct output.

1 answer

Ramr-msft 17,611 Reputation points

2021-09-24T14:30:11.923+00:00

@Swapnil Dave Thanks for the question. Can you please share the sample/Document that you are trying.
SR in ML.NET and SR in Python: https://github.com/microsoft/anomalydetector
Try out the service at Azure Notebook: https://aka.ms/adnotebook
Here is link to the document using ML.NET.
Please sign in to rate this answer.
Swapnil Dave 41 Reputation points

2021-09-25T06:41:39.167+00:00

135097-srcnn.txt

@Ramr-msft
This is the file pretty much the gist of what i am trying to do..

Actually i am a fresher in this ML domain and trying really hard to gain knowledge. I have the task of detecting anomaly in the time series. I have been told that there are two types of anomaly detection, namely : 1] basic 2] agile (like seasonal maybe)

And, also i have been told about that ML Dotnet is good for that. And i have been trying to get a grasp on the way ML code is done in c# dotnet and all the algorithm it provides. I have seen the algorithms for the anomaly detection but i can't understand what these algorithms perform underneath. As of now the computations are just a black box to me, but if provided the right understanding i might be able to perform well. I understood the pipeline flow of ML Context and all but the thing that concerns me is the working of the algorithm. And if i don't know how the algorithm performs i won't be able to pass the required arguments. In a way that they are intended to.

I would love to learn and gain the knowledge about the algorithms if you could teach me. I don't have any teacher who could teach me this.

P.S: I have passed random arguments in the SrCnn Anomaly model for the windows size, training size, threshold and all. Was just checking
Sign in to comment