def test_spark_ndcg(spark_data, target_metrics):
df_true, df_pred = spark_data
evaluator0 = SparkRankingEvaluation(df_true, df_true, k=10, col_prediction="rating")
assert evaluator0.ndcg_at_k() == 1.0
evaluator = SparkRankingEvaluation(df_true, df_pred, k=10)
assert evaluator.ndcg_at_k() == target_metrics["ndcg"]
evaluator1 = SparkRankingEvaluation(
df_true, df_pred, relevancy_method="by_threshold", threshold=3.5
)
assert evaluator1.ndcg_at_k() == target_metrics["ndcg"]
def ranking_metrics_pyspark(test, predictions, k=DEFAULT_K):
rank_eval = SparkRankingEvaluation(
test, predictions, k=k, relevancy_method="top_k", **COL_DICT
)
return {
"MAP": rank_eval.map_at_k(),
"nDCG@k": rank_eval.ndcg_at_k(),
"Precision@k": rank_eval.precision_at_k(),
"Recall@k": rank_eval.recall_at_k(),
}
def ranking_metrics_pyspark(test, predictions, k=DEFAULT_K):
rank_eval = SparkRankingEvaluation(test,
predictions,
k=k,
relevancy_method="top_k",
**COL_DICT)
return {
"MAP": rank_eval.map_at_k(),
"nDCG@k": rank_eval.ndcg_at_k(),
"Precision@k": rank_eval.precision_at_k(),
"Recall@k": rank_eval.recall_at_k()
}
def test_spark_python_match(python_data, spark):
# Test on the original data with k = 10.
df_true, df_pred = python_data
dfs_true = spark.createDataFrame(df_true)
dfs_pred = spark.createDataFrame(df_pred)
eval_spark1 = SparkRankingEvaluation(dfs_true, dfs_pred, k=10)
match1 = [
recall_at_k(df_true, df_pred,
k=10) == pytest.approx(eval_spark1.recall_at_k(), TOL),
precision_at_k(df_true, df_pred,
k=10) == pytest.approx(eval_spark1.precision_at_k(),
TOL),
ndcg_at_k(df_true, df_pred,
k=10) == pytest.approx(eval_spark1.ndcg_at_k(), TOL),
map_at_k(df_true, df_pred,
k=10) == pytest.approx(eval_spark1.map_at_k(), TOL),
]
assert all(match1)
# Test on the original data with k = 3.
dfs_true = spark.createDataFrame(df_true)
dfs_pred = spark.createDataFrame(df_pred)
eval_spark2 = SparkRankingEvaluation(dfs_true, dfs_pred, k=3)
match2 = [
recall_at_k(df_true, df_pred,
k=3) == pytest.approx(eval_spark2.recall_at_k(), TOL),
precision_at_k(df_true, df_pred,
k=3) == pytest.approx(eval_spark2.precision_at_k(),
TOL),
ndcg_at_k(df_true, df_pred,
k=3) == pytest.approx(eval_spark2.ndcg_at_k(), TOL),
map_at_k(df_true, df_pred,
k=3) == pytest.approx(eval_spark2.map_at_k(), TOL),
]
assert all(match2)
# Remove the first row from the original data.
df_pred = df_pred[1:-1]
dfs_true = spark.createDataFrame(df_true)
dfs_pred = spark.createDataFrame(df_pred)
eval_spark3 = SparkRankingEvaluation(dfs_true, dfs_pred, k=10)
match3 = [
recall_at_k(df_true, df_pred,
k=10) == pytest.approx(eval_spark3.recall_at_k(), TOL),
precision_at_k(df_true, df_pred,
k=10) == pytest.approx(eval_spark3.precision_at_k(),
TOL),
ndcg_at_k(df_true, df_pred,
k=10) == pytest.approx(eval_spark3.ndcg_at_k(), TOL),
map_at_k(df_true, df_pred,
k=10) == pytest.approx(eval_spark3.map_at_k(), TOL),
]
assert all(match3)
# Test with one user
df_pred = df_pred[df_pred["userID"] == 3]
df_true = df_true[df_true["userID"] == 3]
dfs_true = spark.createDataFrame(df_true)
dfs_pred = spark.createDataFrame(df_pred)
eval_spark4 = SparkRankingEvaluation(dfs_true, dfs_pred, k=10)
match4 = [
recall_at_k(df_true, df_pred,
k=10) == pytest.approx(eval_spark4.recall_at_k(), TOL),
precision_at_k(df_true, df_pred,
k=10) == pytest.approx(eval_spark4.precision_at_k(),
TOL),
ndcg_at_k(df_true, df_pred,
k=10) == pytest.approx(eval_spark4.ndcg_at_k(), TOL),
map_at_k(df_true, df_pred,
k=10) == pytest.approx(eval_spark4.map_at_k(), TOL),
]
assert all(match4)
def test_spark_python_match(python_data, spark):
# Test on the original data with k = 10.
df_true, df_pred = python_data
dfs_true = spark.createDataFrame(df_true)
dfs_pred = spark.createDataFrame(df_pred)
eval_spark1 = SparkRankingEvaluation(dfs_true, dfs_pred, k=10)
match1 = [
recall_at_k(df_true, df_pred, k=10)
== pytest.approx(eval_spark1.recall_at_k(), TOL),
precision_at_k(df_true, df_pred, k=10)
== pytest.approx(eval_spark1.precision_at_k(), TOL),
ndcg_at_k(df_true, df_pred, k=10)
== pytest.approx(eval_spark1.ndcg_at_k(), TOL),
map_at_k(df_true, df_pred, k=10) == pytest.approx(eval_spark1.map_at_k(), TOL),
]
assert all(match1)
# Test on the original data with k = 3.
dfs_true = spark.createDataFrame(df_true)
dfs_pred = spark.createDataFrame(df_pred)
eval_spark2 = SparkRankingEvaluation(dfs_true, dfs_pred, k=3)
match2 = [
recall_at_k(df_true, df_pred, k=3)
== pytest.approx(eval_spark2.recall_at_k(), TOL),
precision_at_k(df_true, df_pred, k=3)
== pytest.approx(eval_spark2.precision_at_k(), TOL),
ndcg_at_k(df_true, df_pred, k=3) == pytest.approx(eval_spark2.ndcg_at_k(), TOL),
map_at_k(df_true, df_pred, k=3) == pytest.approx(eval_spark2.map_at_k(), TOL),
]
assert all(match2)
# Remove the first row from the original data.
df_pred = df_pred[1:-1]
dfs_true = spark.createDataFrame(df_true)
dfs_pred = spark.createDataFrame(df_pred)
eval_spark3 = SparkRankingEvaluation(dfs_true, dfs_pred, k=10)
match3 = [
recall_at_k(df_true, df_pred, k=10)
== pytest.approx(eval_spark3.recall_at_k(), TOL),
precision_at_k(df_true, df_pred, k=10)
== pytest.approx(eval_spark3.precision_at_k(), TOL),
ndcg_at_k(df_true, df_pred, k=10)
== pytest.approx(eval_spark3.ndcg_at_k(), TOL),
map_at_k(df_true, df_pred, k=10) == pytest.approx(eval_spark3.map_at_k(), TOL),
]
assert all(match3)
# Test with one user
df_pred = df_pred.loc[df_pred["userID"] == 3]
df_true = df_true.loc[df_true["userID"] == 3]
dfs_true = spark.createDataFrame(df_true)
dfs_pred = spark.createDataFrame(df_pred)
eval_spark4 = SparkRankingEvaluation(dfs_true, dfs_pred, k=10)
match4 = [
recall_at_k(df_true, df_pred, k=10)
== pytest.approx(eval_spark4.recall_at_k(), TOL),
precision_at_k(df_true, df_pred, k=10)
== pytest.approx(eval_spark4.precision_at_k(), TOL),
ndcg_at_k(df_true, df_pred, k=10)
== pytest.approx(eval_spark4.ndcg_at_k(), TOL),
map_at_k(df_true, df_pred, k=10) == pytest.approx(eval_spark4.map_at_k(), TOL),
]
assert all(match4)