gudovskiy / contextflow

Official PyTorch code for UAI 2024 paper "ContextFlow++: Generalist-Specialist Flow-based Generative Models with Mixed-variable Context Encoding"
https://arxiv.org/abs/2406.00578
4 stars 1 forks source link

Inference for anomaly detection #2

Open leonardo-holtz-isi opened 1 month ago

leonardo-holtz-isi commented 1 month ago

After the training, how can I run a single prediction for anomaly detection, without any evaluation? I don't quite understand what type of post-processing I need to do for the log prob output to just achieve AD predictions when I don't have GT data

gudovskiy commented 3 weeks ago

@leonardo-holtz-isi AD prediction is simply the likelihood or its normalized version (by softmax/sigmoid) depending on the task.