* Update spacy.scorer, to use P/R/F to support tokenization errors

spacy/scorer.py

@@ -1,78 +1,102 @@
 from __future__ import division
 
 
+class PRFScore(object):
+    """A precision / recall / F score"""
+    def __init__(self):
+        self.tp = 0
+        self.fp = 0
+        self.fn = 0
+
+    def score_set(self, cand, gold):
+        self.tp += len(cand.intersection(gold))
+        self.fp += len(cand - gold)
+        self.fn += len(gold - cand)
+
+    @property
+    def precision(self):
+        return self.tp / (self.tp + self.fp + 1e-100)
+
+    @property
+    def recall(self):
+        return self.tp / (self.tp + self.fn + 1e-100)
+
+    @property
+    def fscore(self):
+        p = self.precision
+        r = self.recall
+        return 2 * ((p * r) / (p + r + 1e-100))
+
+
 class Scorer(object):
     def __init__(self, eval_punct=False):
-        self.heads_corr = 0
+        self.tokens = PRFScore()
-        self.labels_corr = 0
+        self.sbd = PRFScore()
-        self.tags_corr = 0
+        self.unlabelled = PRFScore()
-        self.ents_tp = 0
+        self.labelled = PRFScore()
-        self.ents_fp = 0
+        self.tags = PRFScore()
-        self.ents_fn = 0
+        self.ner = PRFScore()
-        self.total = 1e-100
-        self.mistokened = 0
-        self.n_tokens = 0
         self.eval_punct = eval_punct
 
     @property
     def tags_acc(self):
-        return (self.tags_corr / (self.n_tokens - self.mistokened)) * 100
+        return self.tags.fscore * 100
 
     @property
     def token_acc(self):
-        return (self.mistokened / self.n_tokens) * 100
+        return self.tokens.fscore * 100
-
 
     @property
     def uas(self):
-        return (self.heads_corr / self.total) * 100
+        return self.unlabelled.fscore * 100
 
     @property
     def las(self):
-        return (self.labels_corr / self.total) * 100
+        return self.labelled.fscore * 100
 
     @property
     def ents_p(self):
-        return (self.ents_tp / (self.ents_tp + self.ents_fp + 1e-100)) * 100
+        return self.ner.precision
 
     @property
     def ents_r(self):
-        return (self.ents_tp / (self.ents_tp + self.ents_fn + 1e-100)) * 100
+        return self.ner.recall
 
     @property
     def ents_f(self):
-        return (2 * self.ents_p * self.ents_r) / (self.ents_p + self.ents_r + 1e-100)
+        return self.ner.fscore
 
     def score(self, tokens, gold, verbose=False):
         assert len(tokens) == len(gold)
 
-        for i, token in enumerate(tokens):
+        gold_deps = set()
-            if not self.skip_token(i, token, gold):
+        gold_tags = set()
-                self.total += 1
+        gold_tags = set()
-                if verbose:
+        for id_, word, tag, head, dep, ner in gold.orig_annot:
-                    print token.orth_, token.tag_, token.dep_, token.head.orth_, token.head.i == gold.heads[i]
+            if dep.lower() not in ('p', 'punct'):
-                if token.head.i == gold.heads[i]:
+                gold_deps.add((id_, head, dep))
-                    self.heads_corr += 1
+                gold_tags.add((id_, tag))
-                    self.labels_corr += token.dep_.lower() == gold.labels[i].lower()
+        cand_deps = set()
-            if gold.tags[i] != None:
+        cand_tags = set()
-                self.tags_corr += token.tag_ == gold.tags[i]
+        for token in tokens:
-                self.n_tokens += 1
+            if token.dep_ not in ('p', 'punct') and token.orth_.strip():
-        gold_ents = set((start, end, label) for (start, end, label) in gold.ents)
+                gold_i = gold.cand_to_gold[token.i]
-        guess_ents = set((e.start, e.end, e.label_) for e in tokens.ents)
+                gold_head = gold.cand_to_gold[token.head.i]
-        if verbose and gold_ents:
+                # None is indistinct, so we can't just add it to the set
-            for start, end, label in guess_ents:
+                # Multiple (None, None) deps are possible
-                mark = 'T' if (start, end, label) in gold_ents else 'F'
+                if gold_i is None or gold_head is None:
-                ent_str = ' '.join(tokens[i].orth_ for i in range(start, end))
+                    self.unlabelled.fp += 1
-                print mark, label, ent_str
+                    self.labelled.fp += 1
-            for start, end, label in gold_ents:
+                else:
-                if (start, end, label) not in guess_ents:
+                    cand_deps.add((gold_i, gold_head, token.dep_))
-                    ent_str = ' '.join(tokens[i].orth_ for i in range(start, end))
+                if gold_i is None:
-                    print 'M', label, ent_str
+                    self.tags.fp += 1
-            print
+                else:
-        if gold_ents:
+                    cand_tags.add((gold_i, token.tag_))
-            self.ents_tp += len(gold_ents.intersection(guess_ents))
-            self.ents_fn += len(gold_ents - guess_ents)
-            self.ents_fp += len(guess_ents - gold_ents)
 
-    def skip_token(self, i, token, gold):
+        self.tags.score_set(cand_tags, cand_deps)
-        return gold.labels[i] in ('P', 'punct') or gold.heads[i] == None
+        self.labelled.score_set(cand_deps, gold_deps)
+        self.unlabelled.score_set(
+            set(item[:2] for item in cand_deps),
+            set(item[:2] for item in gold_deps),
+        )