Here we list the pretrained models released along with our SDSS paper. Instructions for use and example notebooks are available in the project github repository. All results in the paper were generated with the models below.
Pretrained self-supervised model:
File path: https://portal.nersc.gov/project/dasrepo/self-supervised-learning-sdss/checkpoints/pretrained_paper_model.pth.tar, 350 MB.
This pytorch model contains weights of the CNN encoder trained via our contrastive self-supervised learning framework.
We recommend using a tool like wget for downloading data. Usage is:
wget https://portal.nersc.gov/project/dasrepo/self-supervised-learning-sdss/checkpoints/pretrained_paper_model.pth.tar
Fine-tuned photo-z model:
File path: https://portal.nersc.gov/project/dasrepo/self-supervised-learning-sdss/checkpoints/photoz_finetuned_model.pth.tar, 183 MB.
This pytorch model contains weights of the CNN encoder after fine-tuning for the task of redshift estimation using all of our spectroscopic redshift labels.
We recommend using a tool like wget for downloading data. Usage is:
wget https://portal.nersc.gov/project/dasrepo/self-supervised-learning-sdss/checkpoints/photoz_finetuned_model.pth.tar
Supervised photo-z baseline model:
File path: https://portal.nersc.gov/project/dasrepo/self-supervised-learning-sdss/checkpoints/photoz_supervised_baseline_model.pth.tar, 183 MB.
This pytorch model contains weights of the baseline fully-supervised CNN, trained for redshift estimation using all of our spectroscopic redshift labels.
We recommend using a tool like wget for downloading data. Usage is:
wget https://portal.nersc.gov/project/dasrepo/self-supervised-learning-sdss/checkpoints/photoz_supervised_baseline_model.pth.tar