Hacker News Reader

Comment by ericd | Hacker News Reader

This is a new visualization layer on PadMapper (an apartment hunting site) which shows the relative frequency of crime in different areas of San Francisco. It's not differentiated between nonviolent crime and violent crime currently (should be at some point). Red is bad, green is good, and nothing means very little crime reported for that area (even better than green within the bounds of SF, simply no data outside).

View on HN

It would be nice if you could filter by the type of crime, not even necessarily violent crime. I might care if someone's selling heroin outside my door, or cars are being stolen, but probably wouldn't care if prostitutes are running about or dogs are unleashed.

Good point. It currently needs to be precomputed as images, so it's hard to make it customizable, but it probably makes sense to make a category of unsavory crimes. I would assume that most people dislike living near the same types of crime.

How are you precomputing the images? I thought GMaps provided a way to input data and get a heatmap, but apparently I'm mistaken.

I couldn't find anything to that effect, so I wrote custom scripts. Basically, it creates an image pixel by pixel, with the pixel value corresponding to the crime density of that chunk of the map, blows it up, smooths it with a Gaussian kernel, and then uses some GIS tools to warp it and chop it up into tiles for different zoom levels.

They do if you're using Fusion Tables: http://code.google.com/apis/maps/documentation/javascript/ov...

Personally, I'd be ok with pot and coke, but not heroin or meth.

Also, preferably decent-looking prostitutes, not any of those skeezy ones you see outside the 16th street BART.

Yeah, I think this needs a drugs checkbox selector widget, just like padmapper itself.

That's an interesting feature for an apartment hunting site. However I think you have to add some information like the time-interval this goes back to. If it's only for a couple of days the statistic might be full of relative false-positives and false-negatives.

Ah yeah, I should specify that. The data window is the past 58 weeks or so.

Reply on news.ycombinator.com