yzhao062/xgbod

Supplementary material for IJCNN paper "XGBOD: Improving Supervised Outlier Detection with Unsupervised Representation Learning"

Hi there, I'm Yue ZHAO (赵越 in Chinese)! 👋

😄 I am an Assistant Professor at USC Computer Science; see the latest information at my homepage.

Prospective Students.We plan to recruit 1-2 Ph.D. students for Fall 2025. For undergraduate/graduate interns, it will be considered on a case basis. I personally do not want to have a large group of "interns" without enough support. See details at my homepage.

🌱 Research Interests. My work focuses on creating robust, efficient, and automated machine learning (ML) and data mining (DM) algorithms, systems, and applications. My primary areas of interest are:

1. Robustness and Security of AI: Enhancing the robustness and security of AI systems through out-of-distribution (OOD) detection, outlier detection, and anomaly detection.
2. Efficient and Scalable AI: Developing efficient and scalable ML systems and automation techniques.
3. Applications in Security, Finance, and Healthcare: Applying AI technologies to address complex problems in security, finance, and healthcare sectors.

⚡ Open-source Contribution: I created PyOD (used by NASA, Tesla, Morgan Stanley, and more) - the most popular library for anomaly detection in 2017. Also, I have led more than 10 ML open-source initiatives, receiving 20,000 GitHub stars (top 0.002%) and >22M downloads. Popular ones: PyOD, PyGOD, TDC, ADBench

📫 Contact me by:

• Email (yzhao010 [AT] usc.edu)
• Twitter
• 小红书:「微调」
• 知乎:「微调」
• Homepage

---

xgbod's Issues

Can this algorithm use GPU to accelearte trainning?

Can this algorithm use GPU to accelearte trainning? Thank you !

Some errors

This is excellent resource:

I had some errors that I solved to some extend:

1 Error: init() got an unexpected keyword argument 'ratio'

Solution: ratio to sampling_strategy

BalancedBaggingClassifier(base_estimator=clf,
sampling_strategy='auto',
replacement=False)

2 Error: Solver lbfgs supports only 'l2' or 'none' penalties, got l1 penalty.

Solution: LogisticRegression(C=1, penalty="l1", solver='liblinear', max_iter=400)

https://stackoverflow.com/questions/60868629/valueerror-solver-lbfgs-supports-only-l2-or-none-penalties-got-l1-penalty

3 Error: It didn't seem to converge

Solution: max_iter= 400 or more in the logisticRegression

clf_list = [XGBClassifier(), LogisticRegression(C=1, penalty="l1", solver='liblinear', max_iter=400),
LogisticRegression(penalty="l2", max_iter=400)]

https://stackoverflow.com/questions/62658215/convergencewarning-lbfgs-failed-to-converge-status-1-stop-total-no-of-iter