Restore v2 token_acc score implementation (#12073)

In the v3 scorer refactoring, `token_acc` was implemented incorrectly.
It should use `precision` instead of `fscore` for the measure of
correctly aligned tokens / number of predicted tokens.

Fix the docs to reflect that the measure uses the number of predicted
tokens rather than the number of gold tokens.

spacy/scorer.py

@@ -174,7 +174,7 @@ class Scorer:
             prf_score.score_set(pred_spans, gold_spans)
         if len(acc_score) > 0:
             return {
-                "token_acc": acc_score.fscore,
+                "token_acc": acc_score.precision,
                 "token_p": prf_score.precision,
                 "token_r": prf_score.recall,
                 "token_f": prf_score.fscore,

spacy/tests/test_scorer.py

@@ -110,7 +110,7 @@ def test_tokenization(sented_doc):
     )
     example.predicted[1].is_sent_start = False
     scores = scorer.score([example])
-    assert scores["token_acc"] == approx(0.66666666)
+    assert scores["token_acc"] == 0.5
     assert scores["token_p"] == 0.5
     assert scores["token_r"] == approx(0.33333333)
     assert scores["token_f"] == 0.4

website/docs/api/scorer.md

@@ -76,7 +76,7 @@ core pipeline components, the individual score names start with the `Token` or
 
 Scores the tokenization:
 
-- `token_acc`: number of correct tokens / number of gold tokens
+- `token_acc`: number of correct tokens / number of predicted tokens
 - `token_p`, `token_r`, `token_f`: precision, recall and F-score for token
   character spans