Update textcat scorer threshold behavior (#11696)

* Update textcat scorer threshold behavior

For `textcat` (with exclusive classes) the scorer should always use a
threshold of 0.0 because there should be one predicted label per doc and
the numeric score for that particular label should not matter.

* Rename to test_textcat_multilabel_threshold

* Remove all uses of threshold for multi_label=False

* Update Scorer.score_cats API docs

* Add tests for score_cats with thresholds

* Update textcat API docs

* Fix types

* Convert threshold back to float

* Fix threshold type in docstring

* Improve formatting in Scorer API docs

spacy/pipeline/textcat.py

@@ -72,7 +72,7 @@ subword_features = true
     "textcat",
     assigns=["doc.cats"],
     default_config={
-        "threshold": 0.5,
+        "threshold": 0.0,
         "model": DEFAULT_SINGLE_TEXTCAT_MODEL,
         "scorer": {"@scorers": "spacy.textcat_scorer.v1"},
     },
@@ -144,7 +144,8 @@ class TextCategorizer(TrainablePipe):
         model (thinc.api.Model): The Thinc Model powering the pipeline component.
         name (str): The component instance name, used to add entries to the
             losses during training.
-        threshold (float): Cutoff to consider a prediction "positive".
+        threshold (float): Unused, not needed for single-label (exclusive
+            classes) classification.
         scorer (Optional[Callable]): The scoring method. Defaults to
                 Scorer.score_cats for the attribute "cats".
 
@@ -154,7 +155,7 @@ class TextCategorizer(TrainablePipe):
         self.model = model
         self.name = name
         self._rehearsal_model = None
-        cfg = {"labels": [], "threshold": threshold, "positive_label": None}
+        cfg: Dict[str, Any] = {"labels": [], "threshold": threshold, "positive_label": None}
         self.cfg = dict(cfg)
         self.scorer = scorer
 

spacy/scorer.py

@@ -446,7 +446,7 @@ class Scorer:
         labels (Iterable[str]): The set of possible labels. Defaults to [].
         multi_label (bool): Whether the attribute allows multiple labels.
             Defaults to True. When set to False (exclusive labels), missing
-            gold labels are interpreted as 0.0.
+            gold labels are interpreted as 0.0 and the threshold is set to 0.0.
         positive_label (str): The positive label for a binary task with
             exclusive classes. Defaults to None.
         threshold (float): Cutoff to consider a prediction "positive". Defaults
@@ -471,6 +471,8 @@ class Scorer:
         """
         if threshold is None:
             threshold = 0.5 if multi_label else 0.0
+        if not multi_label:
+            threshold = 0.0
         f_per_type = {label: PRFScore() for label in labels}
         auc_per_type = {label: ROCAUCScore() for label in labels}
         labels = set(labels)
@@ -505,20 +507,18 @@ class Scorer:
                 # Get the highest-scoring for each.
                 pred_label, pred_score = max(pred_cats.items(), key=lambda it: it[1])
                 gold_label, gold_score = max(gold_cats.items(), key=lambda it: it[1])
-                if pred_label == gold_label and pred_score >= threshold:
+                if pred_label == gold_label:
                     f_per_type[pred_label].tp += 1
                 else:
                     f_per_type[gold_label].fn += 1
-                    if pred_score >= threshold:
+                    f_per_type[pred_label].fp += 1
-                        f_per_type[pred_label].fp += 1
             elif gold_cats:
                 gold_label, gold_score = max(gold_cats, key=lambda it: it[1])
                 if gold_score > 0:
                     f_per_type[gold_label].fn += 1
             elif pred_cats:
                 pred_label, pred_score = max(pred_cats.items(), key=lambda it: it[1])
-                if pred_score >= threshold:
+                f_per_type[pred_label].fp += 1
-                    f_per_type[pred_label].fp += 1
         micro_prf = PRFScore()
         for label_prf in f_per_type.values():
             micro_prf.tp += label_prf.tp

spacy/tests/pipeline/test_textcat.py

@@ -823,10 +823,10 @@ def test_textcat_loss(multi_label: bool, expected_loss: float):
     assert loss == expected_loss
 
 
-def test_textcat_threshold():
+def test_textcat_multilabel_threshold():
     # Ensure the scorer can be called with a different threshold
     nlp = English()
-    nlp.add_pipe("textcat")
+    nlp.add_pipe("textcat_multilabel")
 
     train_examples = []
     for text, annotations in TRAIN_DATA_SINGLE_LABEL:
@@ -849,7 +849,7 @@ def test_textcat_threshold():
     )
     pos_f = scores["cats_score"]
     assert scores["cats_f_per_type"]["POSITIVE"]["r"] == 1.0
-    assert pos_f > macro_f
+    assert pos_f >= macro_f
 
 
 def test_textcat_multi_threshold():

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

website/docs/api/scorer.md

@@ -229,16 +229,17 @@ The reported `{attr}_score` depends on the classification properties:
 > print(scores["cats_macro_auc"])
 > ```
 
-| Name             | Description                                                                                                                                        |
+| Name             | Description                                                                                                                                                                                        |
-| ---------------- | -------------------------------------------------------------------------------------------------------------------------------------------------- |
+| ---------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `examples`       | The `Example` objects holding both the predictions and the correct gold-standard annotations. ~~Iterable[Example]~~                                |
+| `examples`       | The `Example` objects holding both the predictions and the correct gold-standard annotations. ~~Iterable[Example]~~                                                                                |
-| `attr`           | The attribute to score. ~~str~~                                                                                                                    |
+| `attr`           | The attribute to score. ~~str~~                                                                                                                                                                    |
-| _keyword-only_   |                                                                                                                                                    |
+| _keyword-only_   |                                                                                                                                                                                                    |
-| `getter`         | Defaults to `getattr`. If provided, `getter(doc, attr)` should return the cats for an individual `Doc`. ~~Callable[[Doc, str], Dict[str, float]]~~ |
+| `getter`         | Defaults to `getattr`. If provided, `getter(doc, attr)` should return the cats for an individual `Doc`. ~~Callable[[Doc, str], Dict[str, float]]~~                                                 |
-| labels           | The set of possible labels. Defaults to `[]`. ~~Iterable[str]~~                                                                                    |
+| labels           | The set of possible labels. Defaults to `[]`. ~~Iterable[str]~~                                                                                                                                    |
-| `multi_label`    | Whether the attribute allows multiple labels. Defaults to `True`. ~~bool~~                                                                         |
+| `multi_label`    | Whether the attribute allows multiple labels. Defaults to `True`. When set to `False` (exclusive labels), missing gold labels are interpreted as `0.0` and the threshold is set to `0.0`. ~~bool~~ |
-| `positive_label` | The positive label for a binary task with exclusive classes. Defaults to `None`. ~~Optional[str]~~                                                 |
+| `positive_label` | The positive label for a binary task with exclusive classes. Defaults to `None`. ~~Optional[str]~~                                                                                                 |
-| **RETURNS**      | A dictionary containing the scores, with inapplicable scores as `None`. ~~Dict[str, Optional[float]]~~                                             |
+| `threshold`      | Cutoff to consider a prediction "positive". Defaults to `0.5` for multi-label, and `0.0` (i.e. whatever's highest scoring) otherwise. ~~float~~                                                    |
+| **RETURNS**      | A dictionary containing the scores, with inapplicable scores as `None`. ~~Dict[str, Optional[float]]~~                                                                                             |
 
 ## Scorer.score_links {#score_links tag="staticmethod" new="3"}
 

website/docs/api/textcategorizer.md

@@ -63,7 +63,6 @@ architectures and their arguments and hyperparameters.
 > ```python
 > from spacy.pipeline.textcat import DEFAULT_SINGLE_TEXTCAT_MODEL
 > config = {
->    "threshold": 0.5,
 >    "model": DEFAULT_SINGLE_TEXTCAT_MODEL,
 > }
 > nlp.add_pipe("textcat", config=config)
@@ -82,7 +81,7 @@ architectures and their arguments and hyperparameters.
 
 | Setting     | Description                                                                                                                                                      |
 | ----------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `threshold` | Cutoff to consider a prediction "positive", relevant when printing accuracy results. ~~float~~                                                                   |
+| `threshold` | Cutoff to consider a prediction "positive", relevant for `textcat_multilabel` when calculating accuracy scores. ~~float~~                                        |
 | `model`     | A model instance that predicts scores for each category. Defaults to [TextCatEnsemble](/api/architectures#TextCatEnsemble). ~~Model[List[Doc], List[Floats2d]]~~ |
 | `scorer`    | The scoring method. Defaults to [`Scorer.score_cats`](/api/scorer#score_cats) for the attribute `"cats"`. ~~Optional[Callable]~~                                 |
 
@@ -123,7 +122,7 @@ shortcut for this and instantiate the component using its string name and
 | `model`        | The Thinc [`Model`](https://thinc.ai/docs/api-model) powering the pipeline component. ~~Model[List[Doc], List[Floats2d]]~~       |
 | `name`         | String name of the component instance. Used to add entries to the `losses` during training. ~~str~~                              |
 | _keyword-only_ |                                                                                                                                  |
-| `threshold`    | Cutoff to consider a prediction "positive", relevant when printing accuracy results. ~~float~~                                   |
+| `threshold`    | Cutoff to consider a prediction "positive", relevant for `textcat_multilabel` when calculating accuracy scores. ~~float~~        |
 | `scorer`       | The scoring method. Defaults to [`Scorer.score_cats`](/api/scorer#score_cats) for the attribute `"cats"`. ~~Optional[Callable]~~ |
 
 ## TextCategorizer.\_\_call\_\_ {#call tag="method"}