def test_ips_ctr(self):
# Test immediate calculation of CTR
metric = BinaryRecoMetrics.CTR(click_column='click', estimation='ips')
actual = pd.DataFrame({
Constants.user_id: [1, 2, 3, 4],
Constants.item_id: [1, 2, 0, 3],
'click': [0, 1, 0, 0]
})
predicted = pd.DataFrame({
Constants.user_id: [1, 2, 3, 4],
Constants.item_id: [1, 2, 2, 3],
'click': [0.8, 0.7, 0.8, 0.7]
})
ctr = metric.get_score(actual, predicted)
self.assertEqual(1, ctr)
# Test accumulated calculation
metric = BinaryRecoMetrics.CTR(click_column='click', estimation='ips')
_, results = metric.get_score(actual,
predicted,
batch_accumulate=True,
return_extended_results=True)
self.assertEqual(1, results['ctr'])
self.assertEqual(4, results['support'])
def test_ctr_on_secondary_col(self):
# Tests calculation of CTR on secondary columns
actual = pd.DataFrame({Constants.user_id: [1, 2, 2, 3, 4],
Constants.item_id: [1, 1, 2, 0, 3],
'click': [0, 1, 1, 0, 0],
'kpi': [1, 0, 1, 0, 0]})
predicted = pd.DataFrame({Constants.user_id: [1, 2, 2, 3, 4],
Constants.item_id: [1, 1, 3, 2, 3],
'click': [0.8, 0.7, 0.6, 0.8, 0.7],
'kpi': [1, 0.9, 1, 0, 0]})
# Make sure the click_column is used for generating the items that are in top k.
# When using the click column, due to k=1, row 3 should be ignored.
# Row 4s don't match so they should also be ignored.
# Select recs by click column, but get_score kpi column
metric = BinaryRecoMetrics.CTR(click_column='click', k=1, value_column='kpi')
results = metric.get_score(actual, predicted, return_extended_results=True)
self.assertEqual(1. / 3, results['ctr'])
self.assertEqual(3, results['support'])
# If the value column was being used for selecting the items that are in top k,
# row 2 of predict would be ignored instead of row 3.
# Since row 2s don't match, the CTR would only be evaluated on two points.
metric = BinaryRecoMetrics.CTR(click_column='kpi', k=1, value_column='kpi')
results = metric.get_score(actual, predicted, return_extended_results=True)
self.assertEqual(1. / 2, results['ctr'])
self.assertEqual(2, results['support'])
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_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)
def test_accumulate_when_no_results_this_batch(self):
metrics = CombinedMetrics(
RankingRecoMetrics.Recall(click_column='click', k=1),
BinaryRecoMetrics.CTR(click_column='click', k=1))
actual = pd.DataFrame({
Constants.user_id: [0],
Constants.item_id: [0],
'click': [True]
})
predicted = pd.DataFrame({
Constants.user_id: [0],
Constants.item_id: [0],
'click': [0]
})
batch_res, acc_res = metrics.get_score(actual,
predicted,
batch_accumulate=True,
return_extended_results=True)
self.assertEqual(2, len(batch_res))
self.assertEqual(2, len(acc_res))
actual = pd.DataFrame({
Constants.user_id: [1],
Constants.item_id: [1],
'click': [True]
})
predicted = pd.DataFrame({
Constants.user_id: [2],
Constants.item_id: [2],
'click': [0]
})
batch_res, acc_res = metrics.get_score(actual,
predicted,
batch_accumulate=True,
return_extended_results=True)
self.assertEqual(2, len(batch_res))
self.assertEqual(2, len(acc_res))
def test_inter_list_diversity_in_use(self):
actual = pd.DataFrame({
Constants.user_id: [0, 0, 1, 2],
Constants.item_id: [4, 5, 1, 4],
'score': [True, True, True, True]
})
predicted = pd.DataFrame({
Constants.user_id: [0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2],
Constants.item_id: [0, 1, 2, 3, 0, 1, 2, 4, 0, 1, 2, 5],
'score':
[0.9, 0.7, 0.6, 0.3, 0.9, 0.7, 0.4, 0.1, 0.9, 0.8, 0.6, 0.6]
})
metrics = CombinedMetrics(
RankingRecoMetrics.Recall(click_column='score', k=4),
BinaryRecoMetrics.CTR(click_column='score', k=4),
DiversityRecoMetrics.InterListDiversity(click_column='score',
k=4,
user_sample_size=2,
num_runs=5))
acc_res = metrics.get_score(actual,
predicted,
batch_accumulate=False,
return_extended_results=True)
self.assertEqual(3, len(acc_res))
self.assertEqual(
0.25, acc_res['Inter-List Diversity@4']['inter-list diversity'])
with self.assertRaises(ValueError):
# This should fail when `batch_accumulate=True`, and `InterListDiversity` gets applied in combined metrics.
batch_res, acc_res = metrics.get_score(
actual,
predicted,
batch_accumulate=True,
return_extended_results=True)