manually expand error bars on bad point (s)

[davidwhogg]

Take one quasar, and take the huge outlier point or two and expand the error bar so it is only 1.5-ish sigma away from the mean. Then run on this hand-adjusted data set.

This issue can only be closed when the procedure is also described in the paper.

[mykytyn]

587730845813965270

[davidwhogg]

once you have done this, open an issue to automate this.

[mykytyn]

Haven't run the optimization yet on this, but is this what you want to see in terms of bad points removal. Right now, I'm just removing the points, but I can change to expand the error bar instead, not sure which is better. http://broiler.astrometry.net/~dwm261/Quasars/587730845813965270/bad_data_test-data.png http://broiler.astrometry.net/~dwm261/Quasars/587730845813965270/bad_data_test-data-mod.png

[mykytyn]

http://broiler.astrometry.net/~dwm261/Quasars/587730845813965270/bad_data_test-data-mod2.png took out a couple more points that were bothering me...

[mykytyn]

http://broiler.astrometry.net/~dwm261/Quasars/587730845813965270/bad_data_test-triangle.png

I ran with the "bad points" removed and this is the resulting triangle plot. Note that tau is now better than it was, but I still think it's too low? (It's about e^-1 on average) (days)

[davidwhogg]

Can you write down the criterion by which you removed points? Let's simplify it to a one-liner...

[mykytyn]

Unfortunately, my criteria was just that a point "looks bad", which in my mind was if there was a cluster of points in time and magnitude, and a clear outlier, I removed that outlier. I'm not sure how to automate this at all, not even sure where to start. Suggestions?