Create a sparse matrix (interactions matrix) of user -> movie rating from the Netflix data set

[audiodude]

LightFM requires a matrix, where the rows are the users and the columns are the movies. A '1' in a cell represents that the user liked that movie.

[audiodude]

We should consider whether we can or need to serialize this matrix, or if we can just recreate it each time.

[audiodude]

Basic algorithm:

• Create a numpy matrix with the right number of rows (users) and columns (movies)
• For each file (mv_00000X.txt):
  • read the movie id
  • For each line of the file:
  • Remap the user id
  • Update the matrix (skip any movie ratings that aren't 4 or 5)

Finally: save the matrix to disk (pickle)

Some sample code:

n = 0
if id_ not in remap:
  remap[id_] = n
  n += 1

data[remap[id_]][movie_id] = 5

[audiodude]

As discussed a couple of weeks ago, the interactions matrix needs to be a coo_matrix: https://docs.scipy.org/doc/scipy/reference/generated/scipy.sparse.coo_matrix.html