search

AstarVienna / METIS_Simulations

Scripts for simulating METIS with ScopeSim
1 stars 0 forks source link

Reproducibility of FITS files #68

Closed hugobuddel closed 2 weeks ago

hugobuddel commented 2 weeks ago

We have not yet verified that someone else can generate the FITS files and get the same md5sums.

I'm almost there.

Attached are the checksums that I got: checksums.20241112.edfb8332.large.txt

And also a directory listing: ls.output.20241112.edfb8332.large.txt

These files have different checksums:

Interestingly, these all seem to have 'the wrong file size', that is, they are all 32MB, while other files are 8MB.

I'll try to put one of those files somewhere.

Did someone else manage to recreate the FITS files correctly?

hugobuddel commented 2 weeks ago

OK, I see that the larger files are created with NDIT > 1, so they have a BITPIX of -64, while the smaller ones have BITPIX of 8. That is what we would expect yes.

So maybe these larger ones have some floating point errors or so that make them different when I create them. Dunno

JenniferKarr commented 2 weeks ago

I think we saw that last time, before the quanitzation was fixed; the checksums don't work for floating point data.

I'll re-run set with NDIT=1 for all files, which will do for this release. For future releases we should think about this a bit more, as we'll likely need NDIT > 1 as we move towards more instrumentally accurate data contents.

hugobuddel commented 2 weeks ago

We can also keep the NDIT > 1 right? I think a more realistic simulation is more important than having it perfectly reproducible

JenniferKarr commented 2 weeks ago

That can work too - we keep the download as is, and use the checksums as a test of download integrity rather than reproducibility.

JenniferKarr commented 2 weeks ago

I'll close this as having decided to live with the differences.