evaluator.py

import numpy as np
import External_Libraries.Notebooks_utils.evaluation_function as eval

def evaluate(target_users, recommender, URM, at=10):
    cumulative_precision = 0.0
    cumulative_recall = 0.0
    cumulative_MAP = 0.0
    num_eval = 0

    goodUsers = []

    for user in target_users:

        #if num_eval % 10000 == 0:
            #print("Evaluated user {} of {}".format(num_eval, len(target_users)))

        start_pos = URM.indptr[user]
        end_pos = URM.indptr[user + 1]
        relevant_items = np.array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0])
        if end_pos - start_pos > 0:
            relevant_items = URM.indices[start_pos:end_pos]
            is_relevant = np.in1d(recommender.recommend(user, cutoff=at), relevant_items, assume_unique=True)

        else:
            is_relevant = np.array([False, False, False, False, False, False, False, False, False, False])

        if True in is_relevant[:2]:
            goodUsers.append(user)
        num_eval += 1

        cumulative_precision += eval.precision(is_relevant, relevant_items)
        cumulative_recall += eval.recall(is_relevant, relevant_items)
        cumulative_MAP += eval.MAP(is_relevant, relevant_items)

    cumulative_precision /= num_eval
    cumulative_recall /= num_eval
    cumulative_MAP /= num_eval

    print("Recommender performance is: Precision = {:.4f}, Recall = {:.4f}, MAP@10 = {:.4f}".format(
        cumulative_precision, cumulative_recall, cumulative_MAP))

    result_dict = {
        "precision": cumulative_precision,
        "recall": cumulative_recall,
        "MAP": cumulative_MAP,
        "Users": goodUsers
    }
    return result_dict