Yue Zhao

icon indicating a website link https://viterbi-web.usc.edu/~yzhao010/

icon company University of Southern California icon location Los Angeles, CA, USA icon location Assistant Professor of Computer Science @ University of Southern California (USC); Ph.D. @ CMU. Anomaly/Outlier/OOD Detection | ML Systems | AutoML

Results 97 comments of Yue Zhao

copod od result strange. Especially, continouse od result may be wrong

sorry for being late on this. I think the tool idea is cool. if you could build it, we could add it to base class to run it. also wonder...

Does input data not support sparse matrix ?

I think only certain models support that. Most of them are not. So you have to do some conversion first. Will take this down for future development.

When k is too large, MO_GAAL cannot operate normally

Sorry for being late on this. MO and SO GAAL is adapted from their original implementation. PR is welcomed!

Good Tutorial Slides from UIUC

Noted. Will add it.

Metrics for unsupervised learning for various algorithms

Unfortunately unsupervised OD does not have a universal loss or metric. I know works like this one for internal validation: https://core.ac.uk/download/pdf/37525751.pdf But they are not widely used...

Refactor: have consistency name for keras model sub-object

this is possible. could you tell us which algorithm are u referring to?

autoencoder StandardScaler and sigmoid as output layer / inconsistency in values size

autoencoder is more like a type of neural architecture with many design choices. It is not necessary to follow the exact setting. That being said, feel free to modify the...

autoencoder StandardScaler and sigmoid as output layer / inconsistency in values size

I think your concern makes sense. Maybe we should change the sigmoid to relu or leakyrelu. The reason to not use minmax scalaer for input is it is sensitive to...

Inconsistent naming of n_neighbors and checking of arguments in COF

Thanks! Issue acknowledged and welcome for PR :) Please submit the PR to the development branch.

Why models.__init__.py is commented? How to import models from relative path?

This was a legacy issue and we decided to go for absolute import. Maybe a nice relative import should be preferred. Will mark it as an enhancement to implement.