PySSL_SERS

Implementation of PySSL_SERS

Requirements and installation

We will set up the environment using Miniconda. Clone the current repo

git clone https://github.com/junho1ee/pyssl_sers.git

This is an example for how to set up a working conda environment to run the code (but make sure to use the correct pytorch, cuda versions or cpu only versions):

conda create --name pyssl_sers python=3.9
conda activate pyssl_sers
conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia
pip install lightning==2.0.9

Dataset

Download

If you want to train our models with the data then:

1. download it from links:
   • Bacteria-ID
   • COVID19 serum
2. unzip the directory and place it into data such that you have the path data/bacteria-id/org and data/covid/org

Preprocess

If you want to preprocess the data then:

python preprocess/data_preprocess_bacteria.py
python preprocess/data_preprocess_covid.py

Running the code

Pretraining a new model

python lightning_pretrain_ssl.py --pre byol --augtype phys

Finetuning and test a pretrained model

python lightning_finetune_pred.py --pre byol --task class30 --augtype phys --fold 0

Code references

Our implementation referred to thre following repositories:

• PySSL (https://github.com/giakou4/pyssl)
• fairseq-signals (https://github.com/Jwoo5/fairseq-signals)

Contact

If you have any questions, please contact

• [email protected]