Enchanting Netflix Final
5100 programming teams from 185 countries took part in the purely geek competition Netflix Prize with the main prize of one million dollars.
The competition ended yesterday at 18:42:37 UTC, exactly 30 days after one of the applicants achieved the desired result (RMSE is less than 0.8563). The finish turned out to be truly enchanting: immediately after the first challenger a second appeared, a fierce struggle broke out between them. Literally in the last hours, applicants have changed places with each other several times!
Many of us watched with excitement in real timefor the denouement of this exciting contest. As a result, the winner from the second medalist shared only 0.0001 points. The fate of a million dollars was decided only 4 minutes before the deadline - despite the fact that the contest lasted from October 2, 2006. Who said math contests cannot be dramatic?
Netflix is a movie rental company, both online and offline. The core of their business is Cinematch's unique recommendation system. She predicts which films a person may like based on his past ratings of other films. As you know, it’s the recommendation system that is the Grail for modern e-commerce applications.
By the rulesof the Netflix contest, a development team that could improve Cinematch's performance by at least 10% could qualify for the top million dollar prize. To set up competitive algorithms, a set of initial user data was published: these are 100 million ratings (from 1 to 5 points) for 18 thousand films from 480 thousand randomly selected users. To check the results, a set of pairs of users and films was published, and the algorithm should have predicted the grades as accurately as possible, while the average RMSE (root mean squared error) was calculated. She was supposed to at least 10% improve the performance of the Cinematch algorithm (0.9525). That is, it was necessary to show 0.8572 or less.
The competition lasted almost three years. The developers showed the best result, and in the end, several teams of programmers decided to join forces and make a breakthrough. So the BellKor's Pragmatic Chaos team appeared , which on June 26, 2009 finally managed to improve the result by the cherished 10.08%. According to the rules, after that the rest of the participants had 30 days to improve the result. And competitors seized the opportunity. Less than 24 hours before the deadline, The Ensemble team improved the result to 10.09%.
BellKor programmers also did not sleep in a hat. Apparently, they were ready for this, and 24 minutes before the deadline they posted a new application, with the result the same as that of a competitor.
But the Ensemble team also prepared a “two-way”. They answered literally twenty minutes later, that is, four (!) Minutes before the deadline, with a result of 10.10%, and BellKor already had no time left.
Thus, the competition ended with the victory of The Ensemble team. After checking the results, they should get their well-deserved million.
The competition ended yesterday at 18:42:37 UTC, exactly 30 days after one of the applicants achieved the desired result (RMSE is less than 0.8563). The finish turned out to be truly enchanting: immediately after the first challenger a second appeared, a fierce struggle broke out between them. Literally in the last hours, applicants have changed places with each other several times!
Many of us watched with excitement in real timefor the denouement of this exciting contest. As a result, the winner from the second medalist shared only 0.0001 points. The fate of a million dollars was decided only 4 minutes before the deadline - despite the fact that the contest lasted from October 2, 2006. Who said math contests cannot be dramatic?
Netflix is a movie rental company, both online and offline. The core of their business is Cinematch's unique recommendation system. She predicts which films a person may like based on his past ratings of other films. As you know, it’s the recommendation system that is the Grail for modern e-commerce applications.
By the rulesof the Netflix contest, a development team that could improve Cinematch's performance by at least 10% could qualify for the top million dollar prize. To set up competitive algorithms, a set of initial user data was published: these are 100 million ratings (from 1 to 5 points) for 18 thousand films from 480 thousand randomly selected users. To check the results, a set of pairs of users and films was published, and the algorithm should have predicted the grades as accurately as possible, while the average RMSE (root mean squared error) was calculated. She was supposed to at least 10% improve the performance of the Cinematch algorithm (0.9525). That is, it was necessary to show 0.8572 or less.
The competition lasted almost three years. The developers showed the best result, and in the end, several teams of programmers decided to join forces and make a breakthrough. So the BellKor's Pragmatic Chaos team appeared , which on June 26, 2009 finally managed to improve the result by the cherished 10.08%. According to the rules, after that the rest of the participants had 30 days to improve the result. And competitors seized the opportunity. Less than 24 hours before the deadline, The Ensemble team improved the result to 10.09%.
BellKor programmers also did not sleep in a hat. Apparently, they were ready for this, and 24 minutes before the deadline they posted a new application, with the result the same as that of a competitor.
But the Ensemble team also prepared a “two-way”. They answered literally twenty minutes later, that is, four (!) Minutes before the deadline, with a result of 10.10%, and BellKor already had no time left.
Thus, the competition ended with the victory of The Ensemble team. After checking the results, they should get their well-deserved million.