Merge pull request #63 from chakki-works/enhancement/speedUp

Enhancement/speed up

seqeval/metrics/v1.py

@@ -207,7 +207,8 @@ def precision_recall_fscore_support(y_true: List[List[str]],
                                     sample_weight: Optional[List[int]] = None,
                                     zero_division: str = 'warn',
                                     scheme: Optional[Type[Token]] = None,
-                                    suffix: bool = False) -> SCORES:
+                                    suffix: bool = False,
+                                    **kwargs) -> SCORES:
     """Compute precision, recall, F-measure and support for each class.
 
     Args:
@@ -288,9 +289,11 @@ def precision_recall_fscore_support(y_true: List[List[str]],
         modified with ``zero_division``.
     """
     def extract_tp_actual_correct(y_true, y_pred, suffix, scheme):
-        target_names = unique_labels(y_true, y_pred, scheme, suffix)
-        entities_true = Entities(y_true, scheme, suffix)
-        entities_pred = Entities(y_pred, scheme, suffix)
+        # If this function is called from classification_report,
+        # try to reuse entities to optimize the function.
+        entities_true = kwargs.get('entities_true') or Entities(y_true, scheme, suffix)
+        entities_pred = kwargs.get('entities_pred') or Entities(y_pred, scheme, suffix)
+        target_names = sorted(entities_true.unique_tags | entities_pred.unique_tags)
 
         tp_sum = np.array([], dtype=np.int32)
         pred_sum = np.array([], dtype=np.int32)
@@ -376,7 +379,10 @@ def classification_report(y_true: List[List[str]],
 
     if scheme is None or not issubclass(scheme, Token):
         scheme = auto_detect(y_true, suffix)
-    target_names = unique_labels(y_true, y_pred, scheme, suffix)
+
+    entities_true = Entities(y_true, scheme, suffix)
+    entities_pred = Entities(y_pred, scheme, suffix)
+    target_names = sorted(entities_true.unique_tags | entities_pred.unique_tags)
 
     if output_dict:
         reporter = DictReporter()
@@ -393,7 +399,9 @@ def classification_report(y_true: List[List[str]],
         sample_weight=sample_weight,
         zero_division=zero_division,
         scheme=scheme,
-        suffix=suffix
+        suffix=suffix,
+        entities_true=entities_true,
+        entities_pred=entities_pred
     )
     for row in zip(target_names, p, r, f1, s):
         reporter.write(*row)
@@ -408,7 +416,9 @@ def classification_report(y_true: List[List[str]],
             sample_weight=sample_weight,
             zero_division=zero_division,
             scheme=scheme,
-            suffix=suffix
+            suffix=suffix,
+            entities_true=entities_true,
+            entities_pred=entities_pred
         )
         reporter.write('{} avg'.format(average), avg_p, avg_r, avg_f1, support)
     reporter.write_blank()

seqeval/scheme.py

@@ -35,6 +35,9 @@ class Prefix(enum.Flag):
     ANY = I | O | B | E | S | U | L
 
 
+Prefixes = dict(Prefix.__members__)
+
+
 class Tag(enum.Flag):
     SAME = enum.auto()
     DIFF = enum.auto()
@@ -49,25 +52,13 @@ class Token:
 
     def __init__(self, token: str, suffix: bool = False, delimiter: str = '-'):
         self.token = token
-        self.suffix = suffix
-        self.delimiter = delimiter
+        self.prefix = Prefixes[token[-1]] if suffix else Prefixes[token[0]]
+        tag = token[:-1] if suffix else token[1:]
+        self.tag = tag.strip(delimiter) or '_'
 
     def __repr__(self):
         return self.token
 
-    @property
-    def prefix(self):
-        """Extracts a prefix from the token."""
-        prefix = self.token[-1] if self.suffix else self.token[0]
-        return Prefix[prefix]
-
-    @property
-    def tag(self):
-        """Extracts a tag from the token."""
-        tag = self.token[:-1] if self.suffix else self.token[1:]
-        tag = tag.strip(self.delimiter) or '_'
-        return tag
-
     def is_valid(self):
         """Check whether the prefix is allowed or not."""
         if self.prefix not in self.allowed_prefix:
@@ -229,9 +220,9 @@ class Tokens:
 
     def __init__(self, tokens: List[str], scheme: Type[Token],
                  suffix: bool = False, delimiter: str = '-', sent_id: int = None):
-        self.tokens = [scheme(token, suffix=suffix, delimiter=delimiter) for token in tokens]
-        self.scheme = scheme
         self.outside_token = scheme('O', suffix=suffix, delimiter=delimiter)
+        self.tokens = [scheme(token, suffix=suffix, delimiter=delimiter) for token in tokens]
+        self.extended_tokens = self.tokens + [self.outside_token]
         self.sent_id = sent_id
 
     @property
@@ -276,12 +267,6 @@ def _is_end(self, i: int):
         prev = self.extended_tokens[i - 1]
         return token.is_end(prev)
 
-    @property
-    def extended_tokens(self):
-        # append a sentinel.
-        tokens = self.tokens + [self.outside_token]
-        return tokens
-
 
 class Entities:
 
@@ -315,8 +300,8 @@ def auto_detect(sequences: List[List[str]], suffix: bool = False, delimiter: str
     error_message = 'This scheme is not supported: {}'
     for tokens in sequences:
         for token in tokens:
-            token = Token(token, suffix=suffix, delimiter=delimiter)
             try:
+                token = Token(token, suffix=suffix, delimiter=delimiter)
                 prefixes.add(token.prefix)
             except KeyError:
                 raise ValueError(error_message.format(token))

tests/test_scheme.py

@@ -627,22 +627,23 @@ def test_bilou_tokens_without_tag(tokens, expected):
 class TestToken:
 
     def test_raises_type_error_if_input_is_binary_string(self):
-        token = Token('I-組織'.encode('utf-8'))
-        with pytest.raises(TypeError):
-            tag = token.tag
+        with pytest.raises(KeyError):
+            token = Token('I-組織'.encode('utf-8'))
 
     def test_raises_index_error_if_input_is_empty_string(self):
-        token = Token('')
         with pytest.raises(IndexError):
-            prefix = token.prefix
+            token = Token('')
+
+    def test_representation(self):
+        token = Token('B-ORG')
+        assert 'B-ORG' == str(token)
 
 
 class TestIOB2Token:
 
     def test_invalid_prefix(self):
-        token = IOB2('T')
         with pytest.raises(KeyError):
-            prefix = token.prefix
+            token = IOB2('T')
 
 
 @pytest.mark.parametrize(