Skip to content

Commit b318d9d

Browse files
yzhangcsyzhangcs
committed
Minor fix
1 parent 4dd8000 commit b318d9d

File tree

11 files changed

+188
-112
lines changed

11 files changed

+188
-112
lines changed

‎supar/cmds/biaffine_dep.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -9,7 +9,7 @@
99
def main():
1010
parser = argparse.ArgumentParser(description='Create Biaffine Dependency Parser.')
1111
parser.add_argument('--tree', action='store_true', help='whether to ensure well-formedness')
12-
parser.add_argument('--proj', action='store_true', help='whether to projectivise the data')
12+
parser.add_argument('--proj', action='store_true', help='whether to projectivize the data')
1313
parser.add_argument('--partial', action='store_true', help='whether partial annotation is included')
1414
parser.set_defaults(Parser=BiaffineDependencyParser)
1515
subparsers = parser.add_subparsers(title='Commands', dest='mode')

‎supar/models/con.py

Lines changed: 26 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -17,20 +17,27 @@ class CRFConstituencyModel(Model):
1717
Args:
1818
n_words (int):
1919
The size of the word vocabulary.
20-
n_rels (int):
20+
n_labels (int):
2121
The number of labels in the treebank.
2222
n_tags (int):
2323
The number of POS tags, required if POS tag embeddings are used. Default: ``None``.
2424
n_chars (int):
2525
The number of characters, required if character-level representations are used. Default: ``None``.
26+
encoder (str):
27+
Encoder to use.
28+
``'lstm'``: BiLSTM encoder.
29+
``'bert'``: BERT-like pretrained language model (for finetuning), e.g., ``'bert-base-cased'``.
30+
Default: ``'lstm'``.
2631
feat (list[str]):
27-
Additional features to use.
32+
Additional features to use, required if ``encoder='lstm'``.
2833
``'tag'``: POS tag embeddings.
2934
``'char'``: Character-level representations extracted by CharLSTM.
30-
``'bert'``: BERT representations, other pretrained langugae models like XLNet are also feasible.
35+
``'bert'``: BERT representations, other pretrained language models like RoBERTa are also feasible.
3136
Default: [``'char'``].
3237
n_embed (int):
3338
The size of word embeddings. Default: 100.
39+
n_pretrained (int):
40+
The size of pretrained word embeddings. Default: 100.
3441
n_feat_embed (int):
3542
The size of feature representations. Default: 100.
3643
n_char_embed (int):
@@ -40,7 +47,7 @@ class CRFConstituencyModel(Model):
4047
char_pad_index (int):
4148
The index of the padding token in the character vocabulary, required if using CharLSTM. Default: 0.
4249
bert (str):
43-
Specifies which kind of language model to use, e.g., ``'bert-base-cased'`` and ``'xlnet-base-cased'``.
50+
Specifies which kind of language model to use, e.g., ``'bert-base-cased'``.
4451
This is required if ``encoder='bert'`` or using BERT features. The full list can be found in `transformers`_.
4552
Default: ``None``.
4653
n_bert_layers (int):
@@ -86,8 +93,10 @@ def __init__(self,
8693
n_labels,
8794
n_tags=None,
8895
n_chars=None,
96+
encoder='lstm',
8997
feat=['char'],
9098
n_embed=100,
99+
n_pretrained=100,
91100
n_feat_embed=100,
92101
n_char_embed=50,
93102
n_char_hidden=100,
@@ -209,20 +218,27 @@ class VIConstituencyModel(CRFConstituencyModel):
209218
Args:
210219
n_words (int):
211220
The size of the word vocabulary.
212-
n_rels (int):
221+
n_labels (int):
213222
The number of labels in the treebank.
214223
n_tags (int):
215224
The number of POS tags, required if POS tag embeddings are used. Default: ``None``.
216225
n_chars (int):
217226
The number of characters, required if character-level representations are used. Default: ``None``.
227+
encoder (str):
228+
Encoder to use.
229+
``'lstm'``: BiLSTM encoder.
230+
``'bert'``: BERT-like pretrained language model (for finetuning), e.g., ``'bert-base-cased'``.
231+
Default: ``'lstm'``.
218232
feat (list[str]):
219-
Additional features to use.
233+
Additional features to use, required if ``encoder='lstm'``.
220234
``'tag'``: POS tag embeddings.
221235
``'char'``: Character-level representations extracted by CharLSTM.
222-
``'bert'``: BERT representations, other pretrained langugae models like XLNet are also feasible.
236+
``'bert'``: BERT representations, other pretrained language models like RoBERTa are also feasible.
223237
Default: [``'char'``].
224238
n_embed (int):
225239
The size of word embeddings. Default: 100.
240+
n_pretrained (int):
241+
The size of pretrained word embeddings. Default: 100.
226242
n_feat_embed (int):
227243
The size of feature representations. Default: 100.
228244
n_char_embed (int):
@@ -232,7 +248,7 @@ class VIConstituencyModel(CRFConstituencyModel):
232248
char_pad_index (int):
233249
The index of the padding token in the character vocabulary, required if using CharLSTM. Default: 0.
234250
bert (str):
235-
Specifies which kind of language model to use, e.g., ``'bert-base-cased'`` and ``'xlnet-base-cased'``.
251+
Specifies which kind of language model to use, e.g., ``'bert-base-cased'``.
236252
This is required if ``encoder='bert'`` or using BERT features. The full list can be found in `transformers`_.
237253
Default: ``None``.
238254
n_bert_layers (int):
@@ -286,8 +302,10 @@ def __init__(self,
286302
n_labels,
287303
n_tags=None,
288304
n_chars=None,
305+
encoder='lstm',
289306
feat=['char'],
290307
n_embed=100,
308+
n_pretrained=100,
291309
n_feat_embed=100,
292310
n_char_embed=50,
293311
n_char_hidden=100,

‎supar/models/dep.py

Lines changed: 54 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -24,14 +24,21 @@ class BiaffineDependencyModel(Model):
2424
The number of POS tags, required if POS tag embeddings are used. Default: ``None``.
2525
n_chars (int):
2626
The number of characters, required if character-level representations are used. Default: ``None``.
27+
encoder (str):
28+
Encoder to use.
29+
``'lstm'``: BiLSTM encoder.
30+
``'bert'``: BERT-like pretrained language model (for finetuning), e.g., ``'bert-base-cased'``.
31+
Default: ``'lstm'``.
2732
feat (list[str]):
28-
Additional features to use.
33+
Additional features to use, required if ``encoder='lstm'``.
2934
``'tag'``: POS tag embeddings.
3035
``'char'``: Character-level representations extracted by CharLSTM.
31-
``'bert'``: BERT representations, other pretrained langugae models like XLNet are also feasible.
36+
``'bert'``: BERT representations, other pretrained language models like RoBERTa are also feasible.
3237
Default: [``'char'``].
3338
n_embed (int):
3439
The size of word embeddings. Default: 100.
40+
n_pretrained (int):
41+
The size of pretrained word embeddings. Default: 100.
3542
n_feat_embed (int):
3643
The size of feature representations. Default: 100.
3744
n_char_embed (int):
@@ -41,7 +48,7 @@ class BiaffineDependencyModel(Model):
4148
char_pad_index (int):
4249
The index of the padding token in the character vocabulary, required if using CharLSTM. Default: 0.
4350
bert (str):
44-
Specifies which kind of language model to use, e.g., ``'bert-base-cased'`` and ``'xlnet-base-cased'``.
51+
Specifies which kind of language model to use, e.g., ``'bert-base-cased'``.
4552
This is required if ``encoder='bert'`` or using BERT features. The full list can be found in `transformers`_.
4653
Default: ``None``.
4754
n_bert_layers (int):
@@ -55,7 +62,8 @@ class BiaffineDependencyModel(Model):
5562
``first``: take the first subtoken. ``last``: take the last subtoken. ``mean``: take a mean over all.
5663
Default: ``mean``.
5764
bert_pad_index (int):
58-
The index of the padding token in BERT vocabulary, required if ``encoder='bert'`` or using BERT features. Default: 0.
65+
The index of the padding token in BERT vocabulary, required if ``encoder='bert'`` or using BERT features.
66+
Default: 0.
5967
freeze (bool):
6068
If ``True``, freezes BERT parameters, required if using BERT features. Default: ``True``.
6169
embed_dropout (float):
@@ -88,8 +96,10 @@ def __init__(self,
8896
n_rels,
8997
n_tags=None,
9098
n_chars=None,
99+
encoder='lstm',
91100
feat=['char'],
92101
n_embed=100,
102+
n_pretrained=100,
93103
n_feat_embed=100,
94104
n_char_embed=50,
95105
n_char_hidden=100,
@@ -230,14 +240,21 @@ class CRFDependencyModel(BiaffineDependencyModel):
230240
The number of POS tags, required if POS tag embeddings are used. Default: ``None``.
231241
n_chars (int):
232242
The number of characters, required if character-level representations are used. Default: ``None``.
243+
encoder (str):
244+
Encoder to use.
245+
``'lstm'``: BiLSTM encoder.
246+
``'bert'``: BERT-like pretrained language model (for finetuning), e.g., ``'bert-base-cased'``.
247+
Default: ``'lstm'``.
233248
feat (list[str]):
234-
Additional features to use.
249+
Additional features to use, required if ``encoder='lstm'``.
235250
``'tag'``: POS tag embeddings.
236251
``'char'``: Character-level representations extracted by CharLSTM.
237-
``'bert'``: BERT representations, other pretrained langugae models like XLNet are also feasible.
252+
``'bert'``: BERT representations, other pretrained language models like RoBERTa are also feasible.
238253
Default: [``'char'``].
239254
n_embed (int):
240255
The size of word embeddings. Default: 100.
256+
n_pretrained (int):
257+
The size of pretrained word embeddings. Default: 100.
241258
n_feat_embed (int):
242259
The size of feature representations. Default: 100.
243260
n_char_embed (int):
@@ -247,7 +264,7 @@ class CRFDependencyModel(BiaffineDependencyModel):
247264
char_pad_index (int):
248265
The index of the padding token in the character vocabulary, required if using CharLSTM. Default: 0.
249266
bert (str):
250-
Specifies which kind of language model to use, e.g., ``'bert-base-cased'`` and ``'xlnet-base-cased'``.
267+
Specifies which kind of language model to use, e.g., ``'bert-base-cased'``.
251268
This is required if ``encoder='bert'`` or using BERT features. The full list can be found in `transformers`_.
252269
Default: ``None``.
253270
n_bert_layers (int):
@@ -261,7 +278,8 @@ class CRFDependencyModel(BiaffineDependencyModel):
261278
``first``: take the first subtoken. ``last``: take the last subtoken. ``mean``: take a mean over all.
262279
Default: ``mean``.
263280
bert_pad_index (int):
264-
The index of the padding token in BERT vocabulary, required if ``encoder='bert'`` or using BERT features. Default: 0.
281+
The index of the padding token in BERT vocabulary, required if ``encoder='bert'`` or using BERT features.
282+
Default: 0.
265283
freeze (bool):
266284
If ``True``, freezes BERT parameters, required if using BERT features. Default: ``True``.
267285
embed_dropout (float):
@@ -342,14 +360,21 @@ class CRF2oDependencyModel(BiaffineDependencyModel):
342360
The number of POS tags, required if POS tag embeddings are used. Default: ``None``.
343361
n_chars (int):
344362
The number of characters, required if character-level representations are used. Default: ``None``.
363+
encoder (str):
364+
Encoder to use.
365+
``'lstm'``: BiLSTM encoder.
366+
``'bert'``: BERT-like pretrained language model (for finetuning), e.g., ``'bert-base-cased'``.
367+
Default: ``'lstm'``.
345368
feat (list[str]):
346-
Additional features to use.
369+
Additional features to use, required if ``encoder='lstm'``.
347370
``'tag'``: POS tag embeddings.
348371
``'char'``: Character-level representations extracted by CharLSTM.
349-
``'bert'``: BERT representations, other pretrained langugae models like XLNet are also feasible.
372+
``'bert'``: BERT representations, other pretrained language models like RoBERTa are also feasible.
350373
Default: [``'char'``].
351374
n_embed (int):
352375
The size of word embeddings. Default: 100.
376+
n_pretrained (int):
377+
The size of pretrained word embeddings. Default: 100.
353378
n_feat_embed (int):
354379
The size of feature representations. Default: 100.
355380
n_char_embed (int):
@@ -359,7 +384,7 @@ class CRF2oDependencyModel(BiaffineDependencyModel):
359384
char_pad_index (int):
360385
The index of the padding token in the character vocabulary, required if using CharLSTM. Default: 0.
361386
bert (str):
362-
Specifies which kind of language model to use, e.g., ``'bert-base-cased'`` and ``'xlnet-base-cased'``.
387+
Specifies which kind of language model to use, e.g., ``'bert-base-cased'``.
363388
This is required if ``encoder='bert'`` or using BERT features. The full list can be found in `transformers`_.
364389
Default: ``None``.
365390
n_bert_layers (int):
@@ -373,7 +398,8 @@ class CRF2oDependencyModel(BiaffineDependencyModel):
373398
``first``: take the first subtoken. ``last``: take the last subtoken. ``mean``: take a mean over all.
374399
Default: ``mean``.
375400
bert_pad_index (int):
376-
The index of the padding token in BERT vocabulary, required if ``encoder='bert'`` or using BERT features. Default: 0.
401+
The index of the padding token in BERT vocabulary, required if ``encoder='bert'`` or using BERT features.
402+
Default: 0.
377403
freeze (bool):
378404
If ``True``, freezes BERT parameters, required if using BERT features. Default: ``True``.
379405
embed_dropout (float):
@@ -405,8 +431,10 @@ def __init__(self,
405431
n_rels,
406432
n_tags=None,
407433
n_chars=None,
434+
encoder='lstm',
408435
feat=['char'],
409436
n_embed=100,
437+
n_pretrained=100,
410438
n_feat_embed=100,
411439
n_char_embed=50,
412440
n_char_hidden=100,
@@ -571,14 +599,21 @@ class VIDependencyModel(BiaffineDependencyModel):
571599
The number of POS tags, required if POS tag embeddings are used. Default: ``None``.
572600
n_chars (int):
573601
The number of characters, required if character-level representations are used. Default: ``None``.
602+
encoder (str):
603+
Encoder to use.
604+
``'lstm'``: BiLSTM encoder.
605+
``'bert'``: BERT-like pretrained language model (for finetuning), e.g., ``'bert-base-cased'``.
606+
Default: ``'lstm'``.
574607
feat (list[str]):
575-
Additional features to use.
608+
Additional features to use, required if ``encoder='lstm'``.
576609
``'tag'``: POS tag embeddings.
577610
``'char'``: Character-level representations extracted by CharLSTM.
578-
``'bert'``: BERT representations, other pretrained langugae models like XLNet are also feasible.
611+
``'bert'``: BERT representations, other pretrained language models like RoBERTa are also feasible.
579612
Default: [``'char'``].
580613
n_embed (int):
581614
The size of word embeddings. Default: 100.
615+
n_pretrained (int):
616+
The size of pretrained word embeddings. Default: 100.
582617
n_feat_embed (int):
583618
The size of feature representations. Default: 100.
584619
n_char_embed (int):
@@ -588,7 +623,7 @@ class VIDependencyModel(BiaffineDependencyModel):
588623
char_pad_index (int):
589624
The index of the padding token in the character vocabulary, required if using CharLSTM. Default: 0.
590625
bert (str):
591-
Specifies which kind of language model to use, e.g., ``'bert-base-cased'`` and ``'xlnet-base-cased'``.
626+
Specifies which kind of language model to use, e.g., ``'bert-base-cased'``.
592627
This is required if ``encoder='bert'`` or using BERT features. The full list can be found in `transformers`_.
593628
Default: ``None``.
594629
n_bert_layers (int):
@@ -602,7 +637,8 @@ class VIDependencyModel(BiaffineDependencyModel):
602637
``first``: take the first subtoken. ``last``: take the last subtoken. ``mean``: take a mean over all.
603638
Default: ``mean``.
604639
bert_pad_index (int):
605-
The index of the padding token in BERT vocabulary, required if ``encoder='bert'`` or using BERT features. Default: 0.
640+
The index of the padding token in BERT vocabulary, required if ``encoder='bert'`` or using BERT features.
641+
Default: 0.
606642
freeze (bool):
607643
If ``True``, freezes BERT parameters, required if using BERT features. Default: ``True``.
608644
embed_dropout (float):
@@ -643,8 +679,10 @@ def __init__(self,
643679
n_rels,
644680
n_tags=None,
645681
n_chars=None,
682+
encoder='lstm',
646683
feat=['char'],
647684
n_embed=100,
685+
n_pretrained=100,
648686
n_feat_embed=100,
649687
n_char_embed=50,
650688
n_char_hidden=100,

0 commit comments

Comments
 (0)