bpr-opt-07.py

#---------------------
# MAKE RECOMMENDATION
#---------------------

def make_recommendation(user_id=None, num_items=10):
    """Recommend items for a given user given a trained model

    Args:
        user_id (int): The id of the user we want to create recommendations for.

        num_items (int): How many recommendations we want to return.

    Returns:
        recommendations (pandas.DataFrame): DataFrame with num_items artist names and scores

    """

    # Grab our user matrix U
    user_vecs = get_variable(graph, session, 'user_factors')

    # Grab our item matrix V
    item_vecs = get_variable(graph, session, 'item_factors')

    # Grab our item bias
    item_bi = get_variable(graph, session, 'item_bias').reshape(-1)

    # Calculate the score for our user for all items. 
    rec_vector = np.add(user_vecs[user_id, :].dot(item_vecs.T), item_bi)

    # Grab the indices of the top users
    item_idx = np.argsort(rec_vector)[::-1][:num_items]

    # Map the indices to artist names and add to dataframe along with scores.
    artists, scores = [], []

    for idx in item_idx:
        artists.append(item_lookup.artist.loc[item_lookup.artist_id == str(idx)].iloc[0])
        scores.append(rec_vector[idx])

    recommendations = pd.DataFrame({'artist': artists, 'score': scores})

    return recommendations

print(make_recommendation(user_id=0))