Quick start

Once you’ve built and installed pyrecsys, you can:

0. Set VERBOSE mode, to see some messages:

   >>> import recsys.algorithm
   >>> recsys.algorithm.VERBOSE = True
   

1. Load a dataset (first download the Movielens 1M Ratings Data Set, ratings.dat file):

   >>> from recsys.algorithm.factorize import SVD
   >>> svd = SVD()
   >>> svd.load_data(filename='./data/movielens/ratings.dat', sep='::', format={'col':0, 'row':1, 'value':2, 'ids': int})
   Loading ./data/movielens/ratings.dat
   ..........|
   

2. Compute SVD, :

   >>> k = 100
   >>> svd.compute(k=k, min_values=10, pre_normalize=None, mean_center=True, post_normalize=True)
   Creating matrix (1000209 tuples)
   Matrix density is: 4.4684%
   Updating matrix: squish to at least 10 values
   Computing svd k=100, min_values=10, pre_normalize=None, mean_center=True, post_normalize=True
   

you can also save the output SVD model (in a zip file):

>>> k = 100
>>> svd.compute(k=k, min_values=10, pre_normalize=None, mean_center=True, post_normalize=True, savefile='/tmp/movielens')
Creating matrix (1000209 tuples)
Matrix density is: 4.4684%
Updating matrix: squish to at least 10 values
Computing svd k=100, min_values=10, pre_normalize=None, mean_center=True, post_normalize=True
Saving svd model to /tmp/movielens

Note

For more information about svd.compute() parameters see Algorithms section.

once the SVD model has been saved (to a zip file) you can load it anytime, thus there’s not need to svd.compute() it again:

>>> from recsys.algorithm.factorize import SVD
>>> svd2 = SVD(filename='/tmp/movielens') # Loading already computed SVD model
>>> # Get two movies, and compute its similarity:
>>> ITEMID1 = 1    # Toy Story (1995)
>>> ITEMID2 = 2355 # A bug's life (1998)
>>> svd2.similarity(ITEMID1, ITEMID2)
0.67706936677315799

3. Compute similarity between two movies

   >>> ITEMID1 = 1    # Toy Story (1995)
   >>> ITEMID2 = 2355 # A bug's life (1998)
   >>> svd.similarity(ITEMID1, ITEMID2)
   0.67706936677315799
   

4. Get movies similar to Toy Story:

   >>> svd.similar(ITEMID1)
   [(1,    0.99999999999999978), # Toy Story
    (3114, 0.87060391051018071), # Toy Story 2
    (2355, 0.67706936677315799), # A bug's life
    (588,  0.5807351496754426),  # Aladdin
    (595,  0.46031829709743477), # Beauty and the Beast
    (1907, 0.44589398718134365), # Mulan
    (364,  0.42908159895574161), # The Lion King
    (2081, 0.42566581277820803), # The Little Mermaid
    (3396, 0.42474056361935913), # The Muppet Movie
    (2761, 0.40439361857585354)] # The Iron Giant
   

5. Predict rating for a given user and movie,

   >>> MIN_RATING = 0.0
   >>> MAX_RATING = 5.0
   >>> ITEMID = 1
   >>> USERID = 1
   >>> svd.predict(ITEMID, USERID, MIN_RATING, MAX_RATING)
   5.0 #Predicted value
   >>> svd.get_matrix().value(ITEMID, USERID)
   5.0 #Real value
   

6. Recommend (non–rated) movies to a user:

   >>> svd.recommend(USERID, is_row=False) #cols are users and rows are items, thus we set is_row=False
   [(2905, 5.2133848204673416), # Shaggy D.A., The
    (318,  5.2052108435956033), # Shawshank Redemption, The
    (2019, 5.1037438278755474), # Seven Samurai (The Magnificent Seven)
    (1178, 5.0962756861447023), # Paths of Glory (1957)
    (904,  5.0771405690055724), # Rear Window (1954)
    (1250, 5.0744156653222436), # Bridge on the River Kwai, The
    (858,  5.0650911066862907), # Godfather, The
    (922,  5.0605327279819408), # Sunset Blvd.
    (1198, 5.0554543765500419), # Raiders of the Lost Ark
    (1148, 5.0548789542105332)] # Wrong Trousers, The
   

7. Which users should see Toy Story? (e.g. which users -that have not rated Toy Story- would give it a high rating?)

   >>> svd.recommend(ITEMID)
   [(283,  5.716264440514446),
    (3604, 5.6471765418323141),
    (5056, 5.6218800339214496),
    (446,  5.5707524860615738),
    (3902, 5.5494529168484652),
    (4634, 5.51643364021289),
    (3324, 5.5138903299082802),
    (4801, 5.4947999354188548),
    (1131, 5.4941438045650068),
    (2339, 5.4916048051511659)]
   

8. For large datasets (say more than 10M tuples), it might be better to run SVDLIBC directly (divisi2 -that also uses SVDLIBC- is way too slow creating the matrix and computing SVD):

   >>> from recsys.utils.svdlibc import SVDLIBC
   >>> svdlibc = SVDLIBC('./data/movielens/ratings.dat')
   >>> svdlibc.to_sparse_matrix(sep='::', format={'col':0, 'row':1, 'value':2, 'ids': int})
   >>> svdlibc.compute(k=100)
   >>> svd = svdlibc.export()
   >>> svd.similar(ITEMID1) # results might be different than example 4. as there's no min_values=10 set here
   [(1, 0.99999999999999978),
    (3114, 0.84099896392054219),
    (588, 0.79191433686817747),
    (2355, 0.7772760704844065),
    (1265, 0.74946256379033827),
    (364, 0.73730970556786068),
    (2321, 0.73652131961235268),
    (595, 0.71665833726881523),
    (3253, 0.7075696829413568),
    (1923, 0.69687698887991523)]