sample autoencoder implementation

• Implement an autoencoder to reconstruct non anomalous data
• use recon loss thresholding to determine anomalies
  • bonus: use latent embedding space + SVM or scikit's Kernel Density Estimation to determine anomalies

• source NIH (download the train and anomalies)
  • train (uninfected)
  • anomalies (infected)

1. in the root of the project

mkdir data
mkdir logs

2. Download and extract data into a folder data
3. Ready to go! Training and Inference is as follows,

python src/autoencoder/train.py
python src/autoencoder/infer_single.py

soon: will update project to use __main__.py for training and inference

Setup logs directory

python -m pip install -e .

conda create -n autoencoder python=3.10
conda activate autoencoder
python -m pip install -e .