Deep Autoencoding Gaussian Mixture Model for Unsupervised Anomaly Detection (DAGMM)

How to train and evaluate

Train

python src/main.py --mode 'train' --data_path {dataset path}

Test

python main.py --mode 'test_all_point' --data_path {dataset path}

Result Analysis

RUN draw_plot.ipynb

• Qualitative Result: Anomaly score plot for all moment
• Quantative Result: AUROC, AUPRC, Best-F1 Score

Training Details

[Hyperparameter]

Name	Description
Epochs	10
Batch Size	256
Learning Rate	1e-4
lambda_energy	0.1
lambda_cov	0.005
number of gaussian components	5

About DAGMM model

[Paper] Deep Autoencoding Gaussian Mixture Model for Unsupervised Anomaly Detection (ICLR,2018)

[Youtube Review] 발표자: 고려대학교 산업경영공학과 DSBA 연구실 이윤승(https://github.com/yun-ss97)

Reference: [code]