This repo contains the code used to write this medium blog post.
In this blog post, I investigated anomaly detection on the UCSD anomaly detection dataset. With a naive approach and using my physics intuition/python programming knowledge, I investigate some detection algorithms using Deep Learning on a constrained RAM setup (and with no GPU's!). Nothing ground breaking, just a learning experience. And I hope you will enjoy the work as much as I did if you decide to try the algos.
Download https://www.kaggle.com/datasets/karthiknm1/ucsd-anomaly-detection-dataset
Place the content of the zip file in the data folder.
And run one of the notebooks lstm-convolutional-autoencoder.ipynb, lstm-autoencoder-forward.ipynb or lstm-convolutional-autoencoder-forward.ipynb if you wish to train the models or the notebook model-check.ipynb if you wish to check the results yourself.
It is a very easy construct of two terms:
- Anomaly in Japanese is written as 異常, which is pronounced as いじょう (in hiragana) or ijou (in roumaji), and
- the very famous word オタク which is pronounced otaku in roumaji. It is no surprise for those who know me that I consider myself to be a physics otaku, so... I'll let you interpret how and why I chose to name this repo that way. :)