meeting 2023 07 03 - JacobPilawa/TriaxSchwarzschild_wiki_5 GitHub Wiki
- A few diagnostic plots/summary plots from tests I ran over the weekend and in the past few days.
- Note for the jacknife test, I ran 10 realizations, throwing out half the data each time. Individual realizations had something like ~600 points, compared to the fidcuial ~1200 that I had been using up to this point.
First, here's the result of 10 jacknives plotted in our CDF-style summary plot. While it is the case that there is a bit of scatter in the individual jacknife realizations, it doesn't seem like it's a large enough scatter to explain the discrepency between our results and expectations.
| Plot |
|---|
 |
And for a bit more detail, here are the vertical plots for each of the jacknife realizations, showing all the individual recoveries.
| jacknife 0 | jacknife 1 | jacknife 2 | jacknife 3 | jacknife 4 | jacknife 5 | jacknife 6 | jacknife 7 | jacknife 8 | jacknife 9 | |
|---|---|---|---|---|---|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
- After meeting with Emily, I ran another small-ish jacknife test to get a sense of how dependent our results are on the acceptance fraction we impose for the jacknives. These take quite awhile to run, so I was only able to test decreasing the acceptance fraction to ~1/3, and lowering the number of jacknife iterations to n=3. These results are quite robust, though, so I think these results won't change too much with increased n. In general, I don't see anything substantially different from the test above where the acceptance fraction was 1/2 and I ran all realizations (in terms of scatter about a mean value, of course the realizations below individually are different when looking at r1 alone vs r2 alone, but r2 is also in general better recovered). I'm running the rest of the realizations over night now, they just take a bit to run and process!
Click for plots
| r1 only | r2 only | r1 & r2 |
|---|---|---|
 |
 |
 |
I also tried to plot the recoveries of the posteriors in a slightly different way, assuming we want to treat all posteriors on equal footing -- below is a plot of the "normalized" recoveries, (recovered - input)/(uncertainty) for the 150 posteriors, and the distribution for both the fiducial and m_eff case.
| Plot |
|---|
| [[images/230703/posterior_comparisons.png] |