[shortoncash]

You should create geographic bins and sort your tickets into them. Then for each bin, you should identify the number of peaks in the distribution of ticket timestamps associated with a bin. Operating under the assumption that the distribution of timestamps is the sum of multiple distributions (a convolution) each with a well defined mean, each peak represents the mean time when -a- vehicle arrives at your designated geographic bin to assign a ticket.

Now, with all your peaks sorted out for each bin, you want to construct a graph from bin to bin where your graph is constructed where an adjacent node in the graph is added if the numerical distance between the two peaks is the smallest of all choices. You should iteratively repeat the process until all the relevant bins are connected

Then brute force the permutations of arrangements of peaks and you should get a rough approximation of the path of the person issuing tickets.

[davidhegarty]

@shortoncash - I like that approach! If you live up to your username, do you've any interest in doing it as a side project?

[avitalp]

@shortoncash: I like this solution a lot!

@David, are you able to put up a small test batch of the data on GitHub?

[davidhegarty]

Great. We have a blog post up: http://blog.fixed.com/post/116649853902/a-problem-we-need-he...

... I'll setup a GitHub later today.

[davidhegarty]

https://github.com/djburdick/la-street-sweeping-maps

[shortoncash]

I'm booked through August, unfortunately. It's an interesting problem, though.