Tune in, turn on, drop out.

The first hurdle explained in my previous blog was only really semi-solved. In order to get C4 to run, I simply ignored things which didn't have all the information I needed from them. The thing is, I'm not sure I need all the information from them.

This is a redshift dependent problem: source detection in catalogues is run on individual wavebands and colour information is simply crossmatched across the various catalogues. Whilst this is an effective way to detect sources, high redshift objects tend to drop out of shorter wavebands (e.g. u-band, g-band in SDSS) thus the galaxy information lacks these magnitudes. However, what should happen is the flux at that point should be measured so that a meaningful colour measurment can be made. A good reason for doing the former is so that a confirmed object that doesn't appear in certain wavebands will certainly be of some degree of interest. So, high redshift objects will undoubtedly lack this colour information.

So I'm going to make a minimum threshold of 2 colours per object (3 wavebands). I think the trick is to fool C4 into thinking this is a decent way to work and tune the clustering algorithm (I'm resigned to the fact I'm going to have to get into this deep and have a really messy time with it) to consciously decide on clustering the C4 galaxies with limited information. I think I can do this. And if I can do this, I can turn the method round and incorporate more colours. If I can incorporate more colours, I can combine multiple surveys and thus, I have better constraints. And if I do that... well... no-one's ever done that with clusters before.