search

aziz-ayed / denoising

Research to replace the Wavelet approach in the denoising task of the MCCD method by a Machine Learning solution
2 stars 0 forks source link

Make star simulations #1

Open tobias-liaudat opened 3 years ago

tobias-liaudat commented 3 years ago

We need some basic star simulations to test the denoising methods.

This issue is to track the progress of this task. @aziz-ayed

aziz-ayed commented 3 years ago

I added the code for a simulation in the develop branch It is formatted to work in Colab @tobias-liaudat

tobias-liaudat commented 3 years ago

Great, you could add some plots of some stars to check how do they look and verify they are ok.

aziz-ayed commented 3 years ago

Done, however the different stars look a lot like each other Probably we should take a bigger range for the ellipticities ? @tobias-liaudat

tobias-liaudat commented 3 years ago

@aziz-ayed You should check what's the ellipticity of the stars you're plotting. Maybe is that the random ellipticity you're taking is close to zero.

But yes, seeing it again I think we can increase the ranges. We can use [-0.15, 0.15] for both ellipticities. The size range we can leave it as it is.

One thing, as we are not using a lot of stars you shouldn't use np.random.uniform(-0.05, 0.05, 10) but equally spaced values. Something like np.linsapce(start=-0.15, stop=0.15, num=10, endpoint=True). There we can have a regular sampling of the hyper parameter space (e1, e2, R2) and be sure we have all the type of PSFs.

Another thing, maybe we can increase a little the noise. We can go for an SNR of 30.

aziz-ayed commented 3 years ago

Yes, I had plotted several couples of stars to be sure that the lookalike of the ellipticities wasn't due to two random picks having similar values but directly printing the ellipticities sure seems more natural !

Done the changes, also I plotted the first and the thousandth star. As we use linspace and itertools without shuffling, they should be the two stars with the biggest range between their ellipticities.

Now the difference is pretty clear 🙂