def test_reco_quick_start_example(self): # Data actual = pd.DataFrame({ "user_id": [1, 2, 3, 4], "item_id": [1, 2, 0, 3], "clicks": [0, 1, 0, 0] }) predicted = pd.DataFrame({ "user_id": [1, 2, 3, 4], "item_id": [1, 2, 2, 3], "clicks": [0.8, 0.7, 0.8, 0.7] }) # Metrics auc = BinaryRecoMetrics.AUC(click_column="clicks") ctr = BinaryRecoMetrics.CTR(click_column="clicks") ncdg_k = RankingRecoMetrics.NDCG(click_column="clicks", k=3) precision_k = RankingRecoMetrics.Precision(click_column="clicks", k=2) recall_k = RankingRecoMetrics.Recall(click_column="clicks", k=2) map_k = RankingRecoMetrics.MAP(click_column="clicks", k=2) # Scores self.assertEqual(auc.get_score(actual, predicted), 0.25) self.assertEqual(ctr.get_score(actual, predicted), 0.3333333333333333) self.assertEqual(ncdg_k.get_score(actual, predicted), 1) self.assertEqual(precision_k.get_score(actual, predicted), 1) self.assertEqual(recall_k.get_score(actual, predicted), 1) self.assertEqual(map_k.get_score(actual, predicted), 1)
def test_ndcg(self): # First, test the IDCG value idcg_val = idcg(3) self.assertEqual(1. / np.log2(2) + 1. / np.log2(3) + 1. / np.log2(4), idcg_val) # Then, test NDCG # User 1 has items 1, 4, 2 relevant and was recommended items 1, 3, 2 # User 2 checks for no relevant items, shouldn't contribute to the metric # User 3 checks for no recommendations, is 0 # User 4 has items 1, 2 relevant and was recommended items 3, 4; is 0 actual = pd.DataFrame({ Constants.user_id: [1, 1, 1, 1, 3, 4, 4], Constants.item_id: [1, 2, 3, 4, 3, 1, 2], 'click': [1, 1, 0, 1, 1, 1, 1] }) predicted = pd.DataFrame({ Constants.user_id: [1, 1, 1, 2, 4, 4], Constants.item_id: [1, 2, 3, 3, 3, 4], 'click': [0.8, 0.7, 0.75, 0.7, 0.6, 0.4] }) metric = RankingRecoMetrics.NDCG(click_column='click', k=3) results = metric.get_score(actual, predicted, return_extended_results=True) self.assertEqual(((1. / np.log2(2) + 1. / np.log2(4)) / idcg_val) / 3, results['ndcg']) self.assertEqual(3, results['support'])
def test_change_column_names(self): user_id_column = 'uid' item_id_column = 'iid' actual = pd.DataFrame({ user_id_column: [0, 0], item_id_column: [1, 2], 'click': [True, True] }) predicted = pd.DataFrame({ user_id_column: [0, 0, 0, 0], item_id_column: [0, 1, 2, 3], 'click': [0, -1, -2, -3] }) # Test that the output is the same recall_1 = RankingRecoMetrics.Recall('click', k=1, user_id_column=user_id_column, item_id_column=item_id_column) self.assertEqual(0., recall_1.get_score(actual, predicted)) recall_2 = RankingRecoMetrics.Recall('click', k=2, user_id_column=user_id_column, item_id_column=item_id_column) self.assertEqual(0.5, recall_2.get_score(actual, predicted)) # Test that none of the metrics crash metrics = CombinedMetrics( BinaryRecoMetrics.CTR('click', k=1, user_id_column=user_id_column, item_id_column=item_id_column), RankingRecoMetrics.MAP('click', k=1, user_id_column=user_id_column, item_id_column=item_id_column), RankingRecoMetrics.NDCG('click', k=1, user_id_column=user_id_column, item_id_column=item_id_column), RankingRecoMetrics.Precision('click', k=1, user_id_column=user_id_column, item_id_column=item_id_column), RankingRecoMetrics.Recall('click', k=1, user_id_column=user_id_column, item_id_column=item_id_column)) metrics.get_score(actual, predicted)