Yue Zhao

some PyOD models will fail while being used with SUOD

1

The reason is sklearn.clone will lead to issues if the hyperparameters are not well used. Problem can be reproduced by cloning models: This includes: * COD

CNN-based AutoEncoder for Image Outlier Detection

1

The current autoencoder and VAE only supports 2-d data, while there is a need to handle 3-d images. PR welcomed!

should add a test module for each model to see if a model can be safely saved, loaded, and cloned. Simply follow the example here: https://github.com/yzhao062/pyod/blob/master/examples/save_load_model_example.py

You may see this error below  Which is caused by numpy 1.19.4 & windows 10, **you should downgrade to 1.19.3 for now, waiting for numpy for a fix...** **pip...

The dev team of combo supports the PR written in either English or Chinese. However, English is preferred for broader audiences. Other languages may be fine if it is easy...

Support distributed learning and low-precision computation via pytorch lightning

1

We would like to start improving the model scalability via Pytorch lightning (https://www.pytorchlightning.ai/)

We are cleaning up these scripts for an easy run, while the primary results are reproducible with the compare_real_data.py (https://github.com/yzhao062/pytod/tree/main/reproducibility)

implemented local Outlier Probabilites (LoOP)

5

See https://dl.acm.org/citation.cfm?id=1646195