The Napolean Dynamite Problem (2008)

In the end the prize was won and never implemented by Netflix.

http://www.forbes.com/sites/ryanholiday/2012/04/16/what-the-...

> The Napolean Dynamite Problem

1. It's "Napoleon". Copy, paste would have prevented this error.

2. For some reason, an obvious issue is left out of the article -- and that issue is how much of movie ratings constitutes entropy, meaningless noise from which no pattern can be extracted, and how much constitutes actual data that an algorithm might accurately model and predict.

Those who doubt the role of entropy should try compressing the output of a pseudorandom number generator. If the PNG is any good, the compressed output will be larger than its input, implying high entropy. In fact, compressing the predictions of Netflix customers would be a good starting point in an algorithm design -- see whether there are any patterns available to exploit.

3. Along with Napoleon Dynamite on the list of hard-to-predict movies, Repo Man and Eraserhead should be considered -- they're also very hard to categorize.