Debug Px measurements using on-the-fly Gaussian skewers

[andreufont]

Sindhu has made a first measurement, here:
fname = '/global/cfs/cdirs/desi/users/sindhu_s/Lya_Px_measurements/gaussian/analysis-0/contaminated/bf3_binned_out_px-zbins_4-thetabins_20_w_res_wp1d.hdf5'

Andreu has started a new branch in cupix to look at them, with a new TestTheory class that knows the truth power (and can be used with the likelihood class).

add_test_theory

You can find a notebook here:
notebooks/development/debug_gauss_skewers.ipynb

[andreufont]

There seems to be something off with the measurements. They should be the same for all theta bins, but this does not seem to be the case, at least for small theta bins:

On large theta bins, it seems a lot better:

Maybe the errorbars are underestimated for this test?

[sindhusatyavolu]

Yes, I expected scatter at low theta because of fewer pairs, but it appears that the errorbars are small, just to make sure, you haven't plotted the standard error, right? Is it just sqrt(diag(cov))? And these plots are not showing the window convolved theory if I am not wrong?

[andreufont]

Above I was plotting sig_Px = np.sqrt(np.diagonal(data.cov_ZAM[iz][it_M]))

Below I use the plot_px function in the likelihood class: