D3CS — Interactive Curve Shifting

What is this ?

• We are COSMOGRAIL, and we participate in the LSST Strong Lens Time Delay Challenge (TDC).
• The data shown here are light curves from the TDC1.
• We have several time delay measurement algorithms, but they do require a rough initial estimate.
• It seems challenging to get such a robust rough estimate automatically...
• ...and also not trivial to decide if a candidate delay is secure (aka "doubtless"), truly plausible or just a spurious signal.
• We plan to develop automatic techniques for these tasks, building upon the information from the present human classification.
• But, at the end of the day, we will probably never trust a delay which is not doubtless to the eye...

Instructions

When you load a curve pair, the inital time shift is set randomly, the initial mag shift is set to get the blue curve just below the red one.
Your job is to shift the blue curve so to get a good match with the red curve, keeping in mind that extrinsic variability (microlensing) is likely to be present.
To do so,

• drag the blue datapoints with your mouse, or move the sliders below the plot,
• zoom in isotropically on interesing features using your mouse wheel,
• individually adjust the time and magnitude axes, using the arrow keys,
• and drag the whole plot around to explore the data.

We do not really care about the magnitude and flux shifts at this stage, feel free to set them as you prefer.
If you get a nice fit, try to evaluate the uncertainty of your delay estimate using the corresponding slider.
This uncertainty is represented visually by the width of the gray bar below the cursor.
The total width of this bar is twice the value of the 1-σ uncertainty, so that the bar covers the 68% confidence region.
In addition, select how confident you are in your estimate, using the radio buttons:

1. secure if you would bet CHF 1000.- against 1.- that your guess is not a catastrophic error,
2. plausible if you get a good fit and see no other obvious solutions,
3. multimodal if your estimate is clearly only one among two or more possible solutions,
4. uninformative if you're quite sure that no one can ever get any time delay out of this data.

Finally,

• enter a short user name,
• leave an optional comment, for instance to highlight very strange data or explain your decisions,
• and save your measurement.

About the interface

• The interface is written using D3.js, hence the name D3CS for D3 Curve Shifting.
• All this is vector graphics (SVG): once you get tired, you can make it larger by zooming in with your browser!
• From our experience it runs smooth on Safari, Firefox, and Chrome.

Saved results

• The columns that we save into our database are:
1. Date and time of submission
2. IP address
3. Username
4. Rung
5. Pair
6. Time shift [day]
7. Time shift uncertainty [day]
8. Mag shift
9. Flux shift [fraction of median flux]
10. Confidence category (see above)
11. Time between page load and submission [s]
12. [Optional : user specified comment]

Advanced features

• One can load a specific curve with (optional) specific presets using an URL like
  index.php?user=Einstein&loadrung=4&loadpair=7&loadlcbtshift=120.0&loadlcbtshifterr=50.0&loadlcbmshift=-2
  That's a nifty way to inspect estimates from other codes.
• We do have an extended version of the "progress overview" page, displaying the individual estimates and checking for disagreements. But of course, we keep this page private, as we do not want to influence your estimates ;-)