tsurubee/banpei

Input data is to small.

Closed this issue · 4 comments

banpei/banpei/sst.py 50行目
Input data is too small.
がよいかと思われます。

You're passing in data which is smaller than your window value (w)

I am terribly sorry for the late confirmation.
I revised the incorrect comment.
Thank you so much!

Hi tsurubee! Quick question related to this issue: what is the "window value" and how do you determine what "w" is appropriate for a given dataset?

Hi, @jessidirk
"w" means the size of the sliding window.
In order to know what a sliding window is, it is necessary to understand the anomaly detection method called SST.
The following papers are helpful for understanding. (Please check Figure 2)

https://ide-research.net/papers/2005_SDM_Ide.pdf

And, it is a difficult problem to decide the appropriate "w" because the appropriate value strongly depends on the target time series data. So, "w" is a hyperparameter that needs to be decided by yourself considering the periodicity of the data.