From 7b139d1bcd19d313db5b0af52dbfbdf8a921496a Mon Sep 17 00:00:00 2001
From: Adriane Boyd <adrianeboyd@gmail.com>
Date: Tue, 25 Oct 2022 15:22:47 +0200
Subject: [PATCH] Add tests for score_cats with thresholds

---
 spacy/tests/test_scorer.py | 47 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 47 insertions(+)

diff --git a/spacy/tests/test_scorer.py b/spacy/tests/test_scorer.py
index 6e15fa2de..b903f1669 100644
--- a/spacy/tests/test_scorer.py
+++ b/spacy/tests/test_scorer.py
@@ -474,3 +474,50 @@ def test_prf_score():
     assert (a.precision, a.recall, a.fscore) == approx(
         (c.precision, c.recall, c.fscore)
     )
+
+
+def test_score_cats(en_tokenizer):
+    text = "some text"
+    gold_doc = en_tokenizer(text)
+    gold_doc.cats = {"POSITIVE": 1.0, "NEGATIVE": 0.0}
+    pred_doc = en_tokenizer(text)
+    pred_doc.cats = {"POSITIVE": 0.75, "NEGATIVE": 0.25}
+    example = Example(pred_doc, gold_doc)
+    # threshold is ignored for multi_label=False
+    scores1 = Scorer.score_cats(
+        [example],
+        "cats",
+        labels=list(gold_doc.cats.keys()),
+        multi_label=False,
+        positive_label="POSITIVE",
+        threshold=0.1,
+    )
+    scores2 = Scorer.score_cats(
+        [example],
+        "cats",
+        labels=list(gold_doc.cats.keys()),
+        multi_label=False,
+        positive_label="POSITIVE",
+        threshold=0.9,
+    )
+    assert scores1["cats_score"] == 1.0
+    assert scores2["cats_score"] == 1.0
+    assert scores1 == scores2
+    # threshold is relevant for multi_label=True
+    scores = Scorer.score_cats(
+        [example],
+        "cats",
+        labels=list(gold_doc.cats.keys()),
+        multi_label=True,
+        threshold=0.9,
+    )
+    assert scores["cats_macro_f"] == 0.0
+    # threshold is relevant for multi_label=True
+    scores = Scorer.score_cats(
+        [example],
+        "cats",
+        labels=list(gold_doc.cats.keys()),
+        multi_label=True,
+        threshold=0.1,
+    )
+    assert scores["cats_macro_f"] == 0.5