this is for holding javascript data
Melanie added section_Science_Ideas_subsection_Improving__.tex
about 8 years ago
Commit id: eee2ec1692ead597a945a6d1e76d82f378587f4a
deletions | additions
diff --git a/section_Science_Ideas_subsection_Improving__.tex b/section_Science_Ideas_subsection_Improving__.tex
new file mode 100644
index 0000000..0795dcc
--- /dev/null
+++ b/section_Science_Ideas_subsection_Improving__.tex
...
\section{Science Ideas}
\subsection{Improving Simulated Data}
\subsubsection{FERENGI}
What we've done already: Created artificial images of 288 nearby SDSS galaxies spanning the redshift range close to the Hubble galaxies $0.3
What can be improved:
1) In GZH, the Ferengi surface brightness/redshift space did not overlap the Hubble space as completely as we'd like (see Figure~\ref{fig:eyeofsauron}). Selection of a new FERENGI sample should be chosen such that we have at least (50? some number) galaxies in the $z$/$\mu$ bins.
2) In GZH, the galaxies chosen to be FERENGI'd were from SDSS; so the pixel scale needed to be adjusted to match HST. This created a problem for the galaxies at the lowest redshifts, because these images couldn't be redshifted all the way to z = 1.0; so effectively these galaxies provided no useful data. Is there any reason not to start with Hubble galaxies at low z instead, to avoid this problem?
3) Evolution correction - Images were created with a number of assumed evolutions - I'm not sure this was well thought-out, and we should discuss how we could implement this most effectively to the new data.
4) Because of the way the decision tree is structured, we did not have nearly enough answers to higher-order questions beyond 'smooth or features,' so we couldn't use FERENGI to measure redshift bias of users at all. Huge waste. We can improve on this once the new filtering method is implemented: by ensuring potential bars, spirals, clumps, etc are actually seen by users, even if the smoothness of the image would've previously prevented them from ever seeing the question, we can make sure we have enough data to debias *all* questions!
\subsubsection{Illustris}
Illustris provides snapshots of galaxies at a wide range of redshifts, so these could also be used to quantify user bias. The advantage here is that we do not need to bin these in redshift bins - we could more easily fit a continuous function of vote fractions vs redshift, as there are ~60 snapshots of galaxies available between $0
If we do this with Illustris and FERENGI: an interesting comparison could be done by looking at the difference in user bias in the two samples. This might help us see which set is more "realistic" in matching what we'd expect for bias in the "real" (GZH) data.
