lukasruff/deep-sad-pytorch

A PyTorch implementation of Deep SAD, a deep Semi-supervised Anomaly Detection method.

deep-sad-pytorch's Issues

Pretrain

Excuse me. Labeled outlier classes also should be put into the ae_net to train?

AttributeError: 'MyMNIST' object has no attribute 'targets

I faced with an issue with reading dataset:

File "/SVDD/Deep-SAD-PyTorch/src/datasets/mnist.py", line 66, in init
self.semi_targets = torch.zeros_like(self.targets)
AttributeError: 'MyMNIST' object has no attribute 'targets

Any hint?

About parameters for cifar10 experiment

Hi, can you share me corresponding parameters when you train deepSAD and pretrain autoencoder network using cifar10 datasets? I try several parameter sets, but ......., Thank you so much!

Network pretraining issue

Deep-SAD-PyTorch/src/DeepSAD.py, line 123

The encoder has different layer names than the network. As a result, the ae_net_dict here is an empty dictionary. One possible revision,
ae_net_dict = {k: v for k, v in ae_net_dict.items() if 'encoder' in k}

no config file-json

could you please upload the json file for configuration?Thanks!!

Error in pandas setup command

Hello, team! I got a problem when trying to execute the requirement.txt file. There seems syntax issue in pandas package. Not sure if it's caused by pandas 0.24.2 package or by the setup file. Can someone help look into it? Appreciate it!

Supervised Deep Binary Classifier

hello, thanks for your greate work!
could you please release the "Supervised Deep Binary Classifier" code?
Thanks!

Making predictions in testing?

Once I've trained my DeepSAD model, I'd like to make a prediction using the trained model.

In my understanding, when making a prediction for one data, the anomaly score can be calculated using the norm-2 (I assume that the number 2 is omitted) as described in the paper.

Once the anomaly score is calculated, how do I use it to classify a given data?

Thanks in advance.

Training on Custom Dataset

Can you pl guide on steps/details for training the model on new custom dataset ?

Time-series data

Great work! I enjoyed reading your paper.
I am just wondering whether this can be applied to time-series data?

thanks

data generator issue, may affect results.

Deep-SAD-PyTorch/src/datasets/preprocessing.py
In this file, row 45 46, idx_unlabeled_outlier and idx_known_outlier may intersect, since idx_known_outlier_candidates is a subset of idx_outlier.
That will affect the test results.