Question aboout DC2_exp.ipynb

[YicunDuanUMich]

@XinyueLi1012
hi xinyue,

could you please give some instructions on how to reproduce the results in DC2_exp.ipynb? It seems that this file can't be run directly.

yicun

[XinyueLi1012]

Hi Yicun,

When we train with dc2 data, due to memory limit, we split the original image into smaller ones (usually one 4k4k image to 250 8080 images), but for plotting and comparing some prediction results, I used the full-sized image. To do this, I temporarily modify the DC2 class in dc2.py to bypass the splitting (from line 113 to 146) and return the complete image along with the full catalog (and also wcs, matchid).

Here are some other hints about dc2:

1. Make sure that your have set the right set_root_dir to load the catalog using GCRCatalogs package
2. Sometimes to_tile_catalog function in catalog.py won't work for dc2 data (if you want to re-train the model)

[YicunDuanUMich]

Thank you for your reply. I have one more question. Where can we find the matchid? We can't find it in the current dataset. Besides, i will appreciate it if you can provide the path to your training and testing datasets.

[XinyueLi1012]

The matchid could be extracted from the dc2 catalog file (which could be generated using DC2_galaxy_psf_params.ipynb). The dataset I use could be found in /nfs/turbo/lsa-regier/lsstdesc-public/dc2...

[YicunDuanUMich]

Thanks!