popularity.py

"""
Popularity baseline
"""

import numpy as np
from spotlight.factorization._components import _predict_process_ids
from utils import most_common


class PopularityModel(object):

    def __init__(self,
                 k=20):

        self.k = 20
        self.topk = []
        self.flag = False

    def fit(self, interactions, verbose=False):
        self.topk = most_common(interactions.item_ids, 20)
        self.num_items = interactions.num_items

    def predict(self, user_ids, item_ids=None):
        """
        Make predictions: given a user id, compute the recommendation
        scores for items.

        Parameters
        ----------

        user_ids: int or array
           If int, will predict the recommendation scores for this
           user for all items in item_ids. If an array, will predict
           scores for all (user, item) pairs defined by user_ids and
           item_ids.
        item_ids: array, optional
            Array containing the item ids for which prediction scores
            are desired. If not supplied, predictions for all items
            will be computed.

        Returns
        -------

        predictions: np.array
            Predicted scores for all items in item_ids.
        """
        if not self.flag:
            user_ids, item_ids = _predict_process_ids(user_ids, item_ids,
                                                      self.num_items,
                                                      False)
            outs = []
            for item in item_ids:
                if int(item) in self.topk:
                    outs.append(float(1/(self.topk.index(int(item))+1)))
                else:
                    outs.append(0.0)
            self.topk = np.array(outs)
            self.flag = True

        return self.topk