Creating a sparse matrix with LightFM and print predictions

Question

I'm currently working with a Python library called LightFM. But i'm having some trouble with passing the interactions to the fit() method.

Python version: 3 Library: http://lyst.github.io/lightfm/docs/lightfm.html

The documentation states that i should make an sparse matrix of the following type: interactions (np.float32 coo_matrix of shape [n_users, n_items]) – the matrix

But i can't seem to make it work it always recommends the same...

Updated: When executed it the top_items variable say the following no matter which user it iterates over and not any of the other items (Beef or salad), so it seems like i'm doing something wrong. It outputs: ['Cake' 'Cheese'] everytime

Here is my code:

    import numpy as np
from lightfm.datasets import fetch_movielens
from lightfm import LightFM
from scipy.sparse import coo_matrix
import scipy.sparse as sparse
import scipy

// Users, items
data = [
    [1, 0], 
    [2, 1], 
    [3, 2],
    [4, 3]
]

items = np.array(["Cake", "Cheese", "Beef", "Salad"])

data = coo_matrix(data)

#create model
model = LightFM(loss='warp')
#train model
model.fit(data, epochs=30, num_threads=2)

// Print training data
print(data)

def sample_recommendation(model, data, user_ids):

    #number of users and movies in training data
    n_users, n_items = data.shape

    #generate recommendations for each user we input
    for user_id in user_ids:

        #movies our model predicts they will like
        scores = model.predict(user_id, np.arange(n_items))

        #rank them in order of most liked to least
        top_items = items[np.argsort(-scores)]

        print(top_items)

sample_recommendation(model, data, [1,2])

Answer 1

 data = coo_matrix(data)

probably isn't what you want; it's an exact replica of data. Not particularly sparse.

What does data represent?

I'm going to guess that you really want a matrix with mostly 0s, and 1s at the coordinates represented by data.

In [20]: data = [
    ...:     [1, 0], 
    ...:     [2, 1], 
    ...:     [3, 2],
    ...:     [4, 3]
    ...: ]

probably not what you want:

In [21]: ds = sparse.coo_matrix(data)
In [22]: ds.A
Out[22]: 
array([[1, 0],
       [2, 1],
       [3, 2],
       [4, 3]])

try again:

In [23]: data=np.array(data)
In [24]: ds=sparse.coo_matrix((np.ones(4,int),(data[:,0],data[:,1])))
In [25]: ds
Out[25]: 
<5x4 sparse matrix of type '<class 'numpy.int32'>'
    with 4 stored elements in COOrdinate format>
In [26]: ds.A
Out[26]: 
array([[0, 0, 0, 0],
       [1, 0, 0, 0],
       [0, 1, 0, 0],
       [0, 0, 1, 0],
       [0, 0, 0, 1]])

That's more typical of what goes into learning functions.