NetflixPrize.jl

Julia package for handling the Netflix Prize data set of 2006

This package does NOT provide the actual data itself. However, you may download it elsewhere, e.g. from Academic Torrents. Please note that the data set itself comes with a separate license agreement.

1. Place the downloaded training set tarball nf_prize_dataset.tar.gz in ~/Downloads or in the data/ subdirectory under the package name.

2. (Optional but recommended) Fire up some Julia workers on the current node, e.g.

   addprocs(4)

   These extra workers will be used in the next step to speed up data processing.

3. Load the package:

   @everywhere using NetflixPrize

   If you are not using multiple workers, just run

   using NetflixPrize

4. To return the data set as a sparse matrix, run

   NetflixPrize.training_set()

   Where needed, the function will copy the tarball into the package subtree, unpack the tarball, parse all the text files belonging to the training set, and save the resulting sparse matrix in a local JLD (Julia data) file.

   The output is a sparse matrix containing ratings, with rows indexed by movie ID and columns indexed by user ID. (Note: the raw data also contains dates, which are not saved.) Parsing the entire training set can take some time.

   17770x2649429 sparse matrix with 100480507 UInt8 entries:
           [30     ,       6]  =  0x03
   ...
   

Once the file data/training_set.jld has been created, the original tarball and its unpacked contents can be deleted from data/ without adversely affecting the functionality of this package.

The Netflix Prize data set can be cited by

James Bennett, Charles Elkan, Bing Liu, Padhraic Smyth, and Domonkos Tikk, "KDD Cup and Workshop 2007", ACM SiGKDD Explorations Newletter, Vol 9, Iss 2, Dec 2007, pp. 51-52, doi:10.1145/1345448.1345459