Figure out overlapping frames - what they are and how to combine them

[kstoreyf]

see https://docs.google.com/presentation/d/1LOBqTZBd1ACW3gVC4aTnYxSntBUuSiuK4QCtOLID-is/edit#slide=id.p and @joshwalawender's comment on the slide

when combining light curves from different units of the same target:

• if the midpoint of a frame (start time + half of exptime) falls within the midpoint extrema of another image sequence, they overlap?
• if multiple frames overlap, we average their fluxes as a starting approximation? and propagate the uncertainties?
• then the new timestamp for this combined flux is the midpoint between the midpoints of the overlapping frames, and this datapoint is added to master light curve?
  • with this method, best to recalculate master light curve each time new overlapping frames are added

[joshwalawender]

This gets to a core question. When we build the master light curve, do we want to average any points at all? If the master light curve is just a bunch of time series data points (one for each image in a PSC for each camera) then we can operate on that.

For example, we might try fitting a model (which could grow to include many instrumental terms we can't predict right now) to the light curve. Alternatively, we could perform some sort of bayesian statistical test for the existence of a transition (transit entry or exit) at various points in time along the curve. Since this analysis method is to be explored, I think record data and try some of these ideas out without averaging any data points.

[wtgee]

I think we need @oguyon to comment.

See also section 3.4 on his (very outdated) page: http://www.naoj.org/staff/guyon/09allskysurvey.web/56photometry.web/content.html

[kstoreyf]

if for now we consider the curve as time series data points, should the times be the start time of the exposure or the midpoint?

or if @oguyon or anyone has other thoughts on combining light curves, would be appreciated

[wtgee]

Typically you go from the mid-point of a transit but I do think we will want to wait for @oguyon for a few things.

On Thu, Jul 21, 2016 at 5:53 AM, kstoreyf notifications@github.com wrote:

if for now we consider the curve as time series data points, should the times be the start time of the exposure or the midpoint?

or if @oguyon https://github.com/oguyon or anyone has other thoughts on combining light curves, would be appreciated

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/panoptes/PIAA/issues/7#issuecomment-234062366, or mute the thread https://github.com/notifications/unsubscribe-auth/AAEUUOmBFaINu-uOlZy-Ptt8p0aniWEYks5qXnzJgaJpZM4JL_ay .

~Wilfred Tyler Gee

[oguyon]

• sorry for the slow response ... long observing run ongoing on Subaru -

As Josh points out, we probably want to keep the raw measurement points all the way into the final step of the detection algorithm averaging is not always the best thing to do - we'll want to remove statistical outliers, and quite possibly store/explore cross-correlations between measurements

I envision that the data product from a single observation on a star would be a set of single exposure flux measurements in 3 colors

On 07/13/2016 03:10 PM, Josh Walawender wrote:

This gets to a core question. When we build the master light curve, do we want to average any points at all? If the master light curve is just a bunch of time series data points (one for each image in a PSC for each camera) then we can operate on that.

For example, we might try fitting a model (which could grow to include many instrumental terms we can't predict right now) to the light curve. Alternatively, we could perform some sort of bayesian statistical test for the existence of a transition (transit entry or exit) at various points in time along the curve. Since this analysis method is to be explored, I think record data and try some of these ideas out without averaging any data points.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/panoptes/PIAA/issues/7#issuecomment-232530464, or mute the thread https://github.com/notifications/unsubscribe/AFVuDahd74zJaLhwTQR5ClqCYAzuNkpyks5qVYyYgaJpZM4JL_ay.

[oguyon]

exposure midpoint time is better as it makes it easier to mix data points taken with different exposure times

as for transit detection, the ideal case is to observe a full transit with a single PANOPTES observation. Most of the time, though, a single observation will only catch the beginning or end of transit, and the goal is to detect, from a single observation, a sharp drop or increase in flux. Merging data from multiple units then becomes equivalent to lining up these "edge detections" into a periodic ON/OFF transit pattern

On 07/20/2016 10:02 AM, Wilfred Tyler Gee wrote:

Typically you go from the mid-point of a transit but I do think we will want to wait for @oguyon for a few things.

On Thu, Jul 21, 2016 at 5:53 AM, kstoreyf notifications@github.com wrote:

if for now we consider the curve as time series data points, should the times be the start time of the exposure or the midpoint?

or if @oguyon https://github.com/oguyon or anyone has other thoughts on combining light curves, would be appreciated

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/panoptes/PIAA/issues/7#issuecomment-234062366, or mute the thread

https://github.com/notifications/unsubscribe-auth/AAEUUOmBFaINu-uOlZy-Ptt8p0aniWEYks5qXnzJgaJpZM4JL_ay .

~Wilfred Tyler Gee

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/panoptes/PIAA/issues/7#issuecomment-234064756, or mute the thread https://github.com/notifications/unsubscribe-auth/AFVuDSV74FWQtFljfXHF8KcuFBDFl3Clks5qXn7SgaJpZM4JL_ay.