DiogenesAnalytics / autoencoder

Python library implementing various autoencoders.
https://colab.research.google.com/github/DiogenesAnalytics/autoencoder/blob/master/notebooks/demo/anomaly_detection.ipynb
MIT License
0 stars 0 forks source link

Feature: Load/Save Reconstruction Error/Threshold Information #25

Closed DiogenesAnalytics closed 10 months ago

DiogenesAnalytics commented 10 months ago

Problem

Sometimes you may want to save/load autoencoder evaluation data from previous instances. For example maybe you want to use the threshold calculated from the train/val dataset as the threshold on a completely new test dataset. This would allow you to see where the new test dataset reconstruction error distribution would tend to appear relative to the train/val dataset threshold (i.e. how "anomalous" this new test dataset actually is.

Solution

Possible API for this:

DiogenesAnalytics commented 10 months ago

Implemented with: 25e3582be230b29096a57de343db46274904b72c