Mask Transferring Test

[alexji]

@maddiecain and @mgull19 are currently testing how well masks transfer from one r-process star to another.

They've exported/applied the masks (with a small code hack to remove unacceptable measurements) and are ranking each line from 1-3.

• 1 = good fit without any adjustment needed
• 2 = close to good fit (e.g. some regions were automasked previously but needed manual masking the second time)
• 3 = bad fit

This thread is for discussion about how well the transfer works and what improvements may need to be made (if any).

[maddiecain]

https://docs.google.com/spreadsheets/d/1fqcgEZ_ZvQkbyRomqnkaIIOrz7wF5tFgB5i8x9D8dNU/edit#gid=0 Here's my results from applying the masks from MP2307 to MP2318. More results coming soon

[maddiecain]

https://docs.google.com/spreadsheets/d/1fqcgEZ_ZvQkbyRomqnkaIIOrz7wF5tFgB5i8x9D8dNU/edit?usp=sharing this should work

[maddiecain]

Here's the link for the version I"m working on now https://docs.google.com/spreadsheets/d/1cvJuGO75m2PBNGt4OJUjLtlEbjqIArn-TuDamwWUkKk/edit?usp=sharing

[alexji]

@maddiecain can you tell us the stellar parameters for these stars? I think they've been chosen to be very similar.

To summarize ratings for MP2307 -> MP2318:

• 1066 lines in list
• 229 lines marked as acceptable
• 211 rank 1
• 16 rank 2
• 2 rank 3

By element (a few typos):

Elem  N   N1  N2  N3
Al      1   1  0  0
Ba II   1   1  0  0
Ca     12  12  0  0
Ca II --  --  -- --
Ce II   9   9  0  0
Cr I    6   6  0  0
Cr II   1   1  0  0
Er II --  --  -- --
Eu II   5   4  1  0
FeI    94  82 11  1
FeII   17  16  1  0
Gd II --  --  -- --
Ho II   1   1  0  0
K     --  --  -- --
Mg      8   7  1  0
Mn      2   2  0  0
ND II   1   1  0  0
Na      2   2  0  0
Nd II  17  17  0  0
Nd Ii --  --  -- --
Ni      6   5  1  0
O     --  --  -- --
Sc    --  --  -- --
ScII    4   4  0  0
Si      1   1  0  0
Sr II   1   1  0  0
Tb II   2   2  0  0
Ti      2   2  0  0
Ti II  36  34  1  1
V II  --  --  -- --
Zn    --  --  -- --

Overall it seems that it works very well on lines that have been selected to be high quality (but note that is only ~28% of the lines in the original line list).

[maddiecain]

Sure, for MP2318 the stellar parameters are 5880/2.10/-2.55/2.65 in the format temperature/logg/metallicity/microturbulence. For MP2307, it's 4745/1.35/-2.82/1.70. MP2318 is on the horizontal branch, hence the weird parameters. A lot of lines couldn't be used because they didn't get selected when "Fit All" was pressed, although many of them were still usable.

[maddiecain]

Here's a link to our final results:
https://docs.google.com/spreadsheets/d/1P6Y6g8cnx5AOTJQPRRSiabDOUCdTBt6C-J1sI-c_Gbc/edit?usp=sharing

The bottom of the document has the proportion of 1's, 2's, and 3's for each test. The first three are the ones I analyzed, and the second three Maude did. That's why there's a difference in the proportion of 1's, 2's, and 3's - we had slightly different standards for how stars were ranked.

[maudegull]

1 was for good, 2 was for okay and 3 was bad.

[andycasey]

So it seems to me that in most cases it was good (and that masks were useful).

How could they be made more useful? When they were just 'okay' or 'bad', was there a common reason?

[maddiecain]

What I noticed was one overall issue that presented itself in many different ways. Basically, what I often noticed in 2s and 3s was that a region automatically masked in blue in the original version would fail to be masked in the transferred version. Some common errors:
A line close to the center that was previously masked in blue wouldn't be masked in the new version, and the program would fit that line instead of the desired line (usually this happened when the fit line had a much bigger EW)
lines (of any size, but often small) wouldn't be auto-masked when they were in the original version and it would shift the fit down lower than normal (this was by far the most common)
Blends often confused the program and it sometimes wouldn't mask nearby lines and instead fit three or so lines as one in places where it automatically masked in the original program (noticed this a lot in lines similar looking to 4077 because the line sizes would change from star to star)

I'll update this if I think of more. Overall, though, I think that even with the few bad fits this would speed things up a ton and be really useful

[maudegull]

Blends were also most of the times an issue, bc the mask would not be very well fitting on the part blended. And I think one or two times for my masking it was bad when I had used it to adapt the fit by masking areas that weren't lines.

[alexji]

I am removing the science verification label since this isn't technically a test of correctness (and so I can close the milestone). However finishing this test is still important.