search

TSKumarage / Python_Autoencoder

0 stars 1 forks source link

Request #1

Open shahlaebrahimi opened 7 years ago

shahlaebrahimi commented 7 years ago

Hello. Sorry, is it possible to ask you about one class training? Thanks in advance. Best regards,

TSKumarage commented 7 years ago

Hi, One class training is the training method where the autoencoder(neural network) is only trained using the normal class data (remove anomaly labeled data from the training dataset).

Best regards.

shahlaebrahimi commented 7 years ago

Hello, First, many thanks for let me ask you? Then, really I tried some algorithms including XGBoost, Adaboost and etc., on my unbalanced data set, which consists of 1169 negative and 38 positive samples. However, they could not identify positive samples well. If I use one class training like one class svm, I don't know how to determine scoring in order to grid search over hyper-parameters because it is unsupervised method. I would appreciate if you could let me know how to solve my problem? Thanks in advance. Best regards,