Chi2 plot fixes

[maindlt]

Chi2 plot:

• Fixed labels (log / non-log and subscript formatting).
• Fixed colorbar overlap with x-axis if only two parameters are varied.
• Added colorbar label $(\chi^2-\chi^2_\mathrm{min})/\sqrt{2n_\mathrm{obs}}$ Note: $2n_\mathrm{obs}$ is calculated via config.get_2n_obs() which returns the sum of $(2n_\mathrm{obs})$ for all kinematic sets, depending on their type: for GaussHermite, n_obs = number_GH * number_spatial_bins; for BayesLOSVD, n_obs = n_LOSVD_bins * number_spatial_bins.

Closes #312

Tested so far with

• Tutorial notebook 3
• test_nnls.py with varying fixed: False parameters in its config file user_test_config_ml.yaml

[boeckera]

I did the tests that Thomas wrote in #380. I honestly do not know if you want to have the plot like this because I would do several things differently (but plotting is very subjective). For example, if you show data in log(quantity) i would make the scale linear and not log? Also I would always show $10^{12}$ instead of 1e12. I would also make the limits the same for the same quantities, when they are shown in y-axis and x-axis, so it is easier to compare. Lastly, there are no tick labels for the black-hole mass (x-axis). Otherwise I didn't encounter any overlapping labels.

[maindlt]

Hi @boeckera, many thanks for testing and your comments! Of course it is wrong to show log quantities on a log scale - excellent point (and embarrassing missing it, it's been like that forever...). In the last commit this should be fixed and the axis limits should be consistent. As the plotted quantities should be smaller than about 200 in most instances, I did not touch the $10^\mathrm{X}$ vs. 1eX issue, though...

If you have time, a re-test is highly appreciated, as are any comments :-)

Cheers, Thomas

[boeckera]

Hello @maindlt,

Except that we show logarithmic things in log10 not log ;) I changed it haha That is how it looks now:

I do not see how any of these values should ever accede 200, unless you for whatever reason decide to sample things like black hole mass or dark matter fraction in linear space (would you ever want to do that though? I guess that is a different question). Also for me, the question of whether you can show things in log or linear space should be independent of whether you sample things in linear or log space. Additionally, what if I want to show e.g. $M{bh}$ but in logarithmic scale instead of $log(M{bh}$). I guess the question is how complicated/flexible you want to make this plotting routine. Do we want the user to have to possibility to make this plot publishable or should it remain "just" a quick check of what is going on.

[maindlt]

Hello @boeckera, outch, thanks for the log10-fixes! Well, these graphs are mainly to get a quick overview of the results rather than to produce publishable plots. The ticks are a good example for this: sometimes they are close or even overlapping with neighboring axes and need some tweaking; people may prefer to plot log quantities as such and not on a linear scale, etc. Nevertheless our plots need to be mathematically correct. A universal solution will need much development time and it is questionable whether we want and have the capacity to go into providing publication-grade plots for every need ;-) In the past the strategy was to provide a minimal set and let users extend/add as they wish. Shall we pick that up in one of our upcoming development meetings...?

[boeckera]

Hi @maindlt,

no I agree with you that it should be minimal. It is too much work and I would then enforce my plotting style on dynamite (it is already itching in my fingers). From what I gather, it is relatively easy to read in the all_models files and then replicate this plot how one wishes. I think we can merge this now?

[maindlt]

Hi @boeckera,

if you have time of course you can give in to the itching, everybody will be happy and grateful for pretty plots! :-)

Alright, I'll fix the changelog conflict and merge then, ok?

Cheers, Thomas

[boeckera]

Hi @maindlt,

maybe if I do not have anything else to do jajaja. Sounds great!

Cheers, Alina