search

GuansongPang / deviation-network

Source code of the KDD19 paper "Deep anomaly detection with deviation networks", weakly/partially supervised anomaly detection, few-shot anomaly detection, semi-supervised anomaly detection
GNU General Public License v3.0
145 stars 56 forks source link

Interpretability of anomaly scores #9

Closed ThisisHubert closed 2 years ago

ThisisHubert commented 2 years ago

In terms of interpretability of the anomaly scores, can we also get the contribution of each feature importances of the datasets (columns) from the Z-Score-based deviation loss ?

ThisisHubert commented 2 years ago

As far as I've known, we have to use framework like shap or lime to get this kind of parameter

GuansongPang commented 2 years ago

As far as I've known, we have to use framework like shap or lime to get this kind of parameter

Hi,

You may refer to the journal extension of the paper at https://github.com/Choubo/deviation-network-image, where you can find how we can get feature importance w.r.t. the deviation-based anomaly scores.

We focus on image data in the journal version, so the feature importance is the importance of pixels for image data.

ThisisHubert commented 2 years ago

I see. Will take a look and will comment more if more explanation is needed. Thank you for the swift response

ThisisHubert commented 2 years ago

Do you have example for tabular data important feature selection ?

Aml-Hassan-Abd-El-hamid commented 1 year ago

Hi @ThisisHubert did you find an example of tabular data's important feature selection ? and if you did, do you mind sharing it?