The described challenge looks like it's different from the actual game in important ways.<p>Obviously there's the different chances for a 2 vs a 4 (used to be 10%, here it's 25%). More importantly, though, it seems that you don't have to slide any tiles for new ones to be spawned.<p>I've only beaten the game once. By far the hardest part (given the strategy I used) was when I was forced to move in a direction I didn't want to, because all of the other directions didn't move tiles and thus didn't generate a new one. Without this, the game seems like it would be a lot easier.
I love hackerrank and checkio. Both are great platforms. I wish that they would integrate with Coursera/udacity so that I could solve the programming challenges in one of these mediums and at the same time get guidance from a lecture.<p>However, if you want to go through some of the courses, you can get some machine learning as well from lectures on programming a self driving car at udacity and implement them at hackerrank.
The score of a game is very much dependent on luck, though.. I've made an AI bot (<a href="https://github.com/helgefmi/c2048" rel="nofollow">https://github.com/helgefmi/c2048</a>) that averages on ~50k score, but its highscore is ~220k. Sometimes it gets 13k. So to get a good score (or your best score) on hackerrank.com, you'd need to resubmit 500 times. At least with the algorithms that I use.