Smelling Dark Matter

I've been using the little free time I have to compete in kaggle's Observing Dark Worlds competition. I have to say that this machine learning competition has given me a new appreciation of what astronomers deal with in their research. It also illustrated why so many of the astronomy majors I knew in college always looked so sleep deprived - I had attributed it to the nocturnal nature of twinkles. 

In the Observing Dark Worlds competition, the goal is to pin point the locations of invisible dark matter halos with only the coordinates of galaxies and their respective ellipticity. Below is a quick visual example of the difficulty of this task (my code for the visualizations can be found here).

This is the visualization of the data from training sky 150.

The two blue circles are the dark halos hiding in training sky 150.

I haven't made much progress in my analysis - in fact I have never been this stumped on any problem. So far I have only been able to determine the location of one dark matter halo exactly and generate a few probable locations of the other one. Below is the visualization of my progress for training sky 150 and training sky 124 from the competition. The red circles are my estimated locations and the blue circles are the true locations of the dark matter halos.

Training Sky 150

Training Sky 124

If it is before the competition deadline and you are interested in my method or code, I am more than happy to win this competition through a team effort! For my regular readers, I will make a detailed posting after the competition deadline.