def test_distributional_coverage(python_diversity_data, target_metrics):
train_df, reco_df = python_diversity_data
evaluator = PythonDiversityEvaluation(
train_df=train_df, reco_df=reco_df, col_user="******", col_item="ItemId"
)
d_coverage = evaluator.distributional_coverage()
assert d_coverage == target_metrics["d_coverage"]
def test_catalog_coverage(python_diversity_data, target_metrics):
train_df, reco_df = python_diversity_data
evaluator = PythonDiversityEvaluation(
train_df=train_df, reco_df=reco_df, col_user="******", col_item="ItemId"
)
c_coverage = evaluator.catalog_coverage()
assert c_coverage == target_metrics["c_coverage"]
def test_serendipity(python_diversity_data, target_metrics):
train_df, reco_df = python_diversity_data
evaluator = PythonDiversityEvaluation(
train_df=train_df,
reco_df=reco_df,
col_user="******",
col_item="ItemId",
col_relevance="Relevance",
)
assert target_metrics["serendipity"] == evaluator.serendipity()
def test_user_diversity(python_diversity_data, target_metrics):
train_df, reco_df = python_diversity_data
evaluator = PythonDiversityEvaluation(
train_df=train_df, reco_df=reco_df, col_user="******", col_item="ItemId"
)
actual = evaluator.user_diversity()
assert_frame_equal(
target_metrics["user_diversity"],
actual,
check_exact=False,
check_less_precise=4,
)
def test_novelty(python_diversity_data, target_metrics):
train_df, reco_df = python_diversity_data
evaluator = PythonDiversityEvaluation(
train_df=train_df, reco_df=reco_df, col_user="******", col_item="ItemId"
)
novelty = evaluator.novelty()
assert target_metrics["novelty"] == novelty
assert novelty >= 0
# Test that novelty is zero when data includes only one item
train_df_new = train_df.loc[train_df["ItemId"] == 3]
reco_df_new = reco_df.loc[reco_df["ItemId"] == 3]
evaluator = PythonDiversityEvaluation(
train_df=train_df_new, reco_df=reco_df_new, col_user="******", col_item="ItemId"
)
assert evaluator.novelty() == 0
def test_item_novelty(python_diversity_data, target_metrics):
train_df, reco_df = python_diversity_data
evaluator = PythonDiversityEvaluation(
train_df=train_df, reco_df=reco_df, col_user="******", col_item="ItemId"
)
actual = evaluator.historical_item_novelty()
assert_frame_equal(
target_metrics["item_novelty"], actual, check_exact=False, check_less_precise=4
)
assert np.all(actual["item_novelty"].values >= 0)
# Test that novelty is zero when data includes only one item
train_df_new = train_df.loc[train_df["ItemId"] == 3]
evaluator = PythonDiversityEvaluation(
train_df=train_df_new, reco_df=reco_df, col_user="******", col_item="ItemId"
)
actual = evaluator.historical_item_novelty()
assert actual["item_novelty"].values[0] == 0
def test_diversity(python_diversity_data, target_metrics):
train_df, reco_df = python_diversity_data
evaluator = PythonDiversityEvaluation(
train_df=train_df, reco_df=reco_df, col_user="******", col_item="ItemId"
)
assert target_metrics["diversity"] == evaluator.diversity()