Modify: requirements.txt
Browse files- gramformer_backup/__init__.py +1 -0
- gramformer_backup/demo.py +30 -0
- gramformer_backup/gramformer.py +127 -0
- requirements.txt +8 -2
- setup.py +19 -0
gramformer_backup/__init__.py
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
from gramformer.gramformer import Gramformer
|
gramformer_backup/demo.py
ADDED
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from gramformer import Gramformer
|
2 |
+
import torch
|
3 |
+
|
4 |
+
def set_seed(seed):
|
5 |
+
torch.manual_seed(seed)
|
6 |
+
if torch.cuda.is_available():
|
7 |
+
torch.cuda.manual_seed_all(seed)
|
8 |
+
|
9 |
+
set_seed(1212)
|
10 |
+
|
11 |
+
|
12 |
+
gf = Gramformer(models = 1, use_gpu=False) # 1=corrector, 2=detector
|
13 |
+
|
14 |
+
influent_sentences = [
|
15 |
+
"Matt like fish",
|
16 |
+
"the collection of letters was original used by the ancient Romans",
|
17 |
+
"We enjoys horror movies",
|
18 |
+
"Anna and Mike is going skiing",
|
19 |
+
"I walk to the store and I bought milk",
|
20 |
+
"We all eat the fish and then made dessert",
|
21 |
+
"I will eat fish for dinner and drank milk",
|
22 |
+
"what be the reason for everyone leave the company",
|
23 |
+
]
|
24 |
+
|
25 |
+
for influent_sentence in influent_sentences:
|
26 |
+
corrected_sentences = gf.correct(influent_sentence, max_candidates=1)
|
27 |
+
print("[Input] ", influent_sentence)
|
28 |
+
for corrected_sentence in corrected_sentences:
|
29 |
+
print("[Correction] ",corrected_sentence)
|
30 |
+
print("-" *100)
|
gramformer_backup/gramformer.py
ADDED
@@ -0,0 +1,127 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
class Gramformer:
|
2 |
+
|
3 |
+
def __init__(self, models=1, use_gpu=False):
|
4 |
+
from transformers import AutoTokenizer
|
5 |
+
from transformers import AutoModelForSeq2SeqLM
|
6 |
+
from lm_scorer.models.auto import AutoLMScorer as LMScorer
|
7 |
+
import errant
|
8 |
+
self.annotator = errant.load('en')
|
9 |
+
|
10 |
+
if use_gpu:
|
11 |
+
device= "cuda:0"
|
12 |
+
else:
|
13 |
+
device = "cpu"
|
14 |
+
batch_size = 1
|
15 |
+
self.scorer = LMScorer.from_pretrained("gpt2", device=device, batch_size=batch_size)
|
16 |
+
self.device = device
|
17 |
+
correction_model_tag = "prithivida/grammar_error_correcter_v1"
|
18 |
+
self.model_loaded = False
|
19 |
+
|
20 |
+
if models == 1:
|
21 |
+
self.correction_tokenizer = AutoTokenizer.from_pretrained(correction_model_tag)
|
22 |
+
self.correction_model = AutoModelForSeq2SeqLM.from_pretrained(correction_model_tag)
|
23 |
+
self.correction_model = self.correction_model.to(device)
|
24 |
+
self.model_loaded = True
|
25 |
+
print("[Gramformer] Grammar error correct/highlight model loaded..")
|
26 |
+
elif models == 2:
|
27 |
+
# TODO
|
28 |
+
print("TO BE IMPLEMENTED!!!")
|
29 |
+
|
30 |
+
def correct(self, input_sentence, max_candidates=1):
|
31 |
+
if self.model_loaded:
|
32 |
+
correction_prefix = "gec: "
|
33 |
+
input_sentence = correction_prefix + input_sentence
|
34 |
+
input_ids = self.correction_tokenizer.encode(input_sentence, return_tensors='pt')
|
35 |
+
input_ids = input_ids.to(self.device)
|
36 |
+
|
37 |
+
preds = self.correction_model.generate(
|
38 |
+
input_ids,
|
39 |
+
do_sample=True,
|
40 |
+
max_length=128,
|
41 |
+
top_k=50,
|
42 |
+
top_p=0.95,
|
43 |
+
early_stopping=True,
|
44 |
+
num_return_sequences=max_candidates)
|
45 |
+
|
46 |
+
corrected = set()
|
47 |
+
for pred in preds:
|
48 |
+
corrected.add(self.correction_tokenizer.decode(pred, skip_special_tokens=True).strip())
|
49 |
+
|
50 |
+
corrected = list(corrected)
|
51 |
+
scores = self.scorer.sentence_score(corrected, log=True)
|
52 |
+
ranked_corrected = [(c,s) for c, s in zip(corrected, scores)]
|
53 |
+
ranked_corrected.sort(key = lambda x:x[1], reverse=True)
|
54 |
+
return ranked_corrected
|
55 |
+
else:
|
56 |
+
print("Model is not loaded")
|
57 |
+
return None
|
58 |
+
|
59 |
+
def highlight(self, orig, cor):
|
60 |
+
edits = self._get_edits(orig, cor)
|
61 |
+
orig_tokens = orig.split()
|
62 |
+
|
63 |
+
ignore_indexes = []
|
64 |
+
|
65 |
+
for edit in edits:
|
66 |
+
edit_type = edit[0]
|
67 |
+
edit_str_start = edit[1]
|
68 |
+
edit_spos = edit[2]
|
69 |
+
edit_epos = edit[3]
|
70 |
+
edit_str_end = edit[4]
|
71 |
+
|
72 |
+
# if no_of_tokens(edit_str_start) > 1 ==> excluding the first token, mark all other tokens for deletion
|
73 |
+
for i in range(edit_spos+1, edit_epos):
|
74 |
+
ignore_indexes.append(i)
|
75 |
+
|
76 |
+
if edit_str_start == "":
|
77 |
+
if edit_spos - 1 >= 0:
|
78 |
+
new_edit_str = orig_tokens[edit_spos - 1]
|
79 |
+
edit_spos -= 1
|
80 |
+
else:
|
81 |
+
new_edit_str = orig_tokens[edit_spos + 1]
|
82 |
+
edit_spos += 1
|
83 |
+
if edit_type == "PUNCT":
|
84 |
+
st = "<a type='" + edit_type + "' edit='" + \
|
85 |
+
edit_str_end + "'>" + new_edit_str + "</a>"
|
86 |
+
else:
|
87 |
+
st = "<a type='" + edit_type + "' edit='" + new_edit_str + \
|
88 |
+
" " + edit_str_end + "'>" + new_edit_str + "</a>"
|
89 |
+
orig_tokens[edit_spos] = st
|
90 |
+
elif edit_str_end == "":
|
91 |
+
st = "<d type='" + edit_type + "' edit=''>" + edit_str_start + "</d>"
|
92 |
+
orig_tokens[edit_spos] = st
|
93 |
+
else:
|
94 |
+
st = "<c type='" + edit_type + "' edit='" + \
|
95 |
+
edit_str_end + "'>" + edit_str_start + "</c>"
|
96 |
+
orig_tokens[edit_spos] = st
|
97 |
+
|
98 |
+
for i in sorted(ignore_indexes, reverse=True):
|
99 |
+
del(orig_tokens[i])
|
100 |
+
|
101 |
+
return(" ".join(orig_tokens))
|
102 |
+
|
103 |
+
def detect(self, input_sentence):
|
104 |
+
# TO BE IMPLEMENTED
|
105 |
+
pass
|
106 |
+
|
107 |
+
def _get_edits(self, orig, cor):
|
108 |
+
orig = self.annotator.parse(orig)
|
109 |
+
cor = self.annotator.parse(cor)
|
110 |
+
alignment = self.annotator.align(orig, cor)
|
111 |
+
edits = self.annotator.merge(alignment)
|
112 |
+
|
113 |
+
if len(edits) == 0:
|
114 |
+
return []
|
115 |
+
|
116 |
+
edit_annotations = []
|
117 |
+
for e in edits:
|
118 |
+
e = self.annotator.classify(e)
|
119 |
+
edit_annotations.append((e.type[2:], e.o_str, e.o_start, e.o_end, e.c_str, e.c_start, e.c_end))
|
120 |
+
|
121 |
+
if len(edit_annotations) > 0:
|
122 |
+
return edit_annotations
|
123 |
+
else:
|
124 |
+
return []
|
125 |
+
|
126 |
+
def get_edits(self, orig, cor):
|
127 |
+
return self._get_edits(orig, cor)
|
requirements.txt
CHANGED
@@ -1,4 +1,10 @@
|
|
1 |
-
pip==20.1.1
|
2 |
st-annotated-text
|
3 |
beautifulsoup4
|
4 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
st-annotated-text
|
2 |
beautifulsoup4
|
3 |
+
transformers
|
4 |
+
sentencepiece==0.1.95
|
5 |
+
python-Levenshtein==0.12.2
|
6 |
+
fuzzywuzzy==0.18.0
|
7 |
+
tokenizers==0.10.2
|
8 |
+
fsspec==2021.5.0
|
9 |
+
lm-scorer==0.4.2
|
10 |
+
errant
|
setup.py
ADDED
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import setuptools
|
2 |
+
|
3 |
+
setuptools.setup(
|
4 |
+
name="gramformer",
|
5 |
+
version="1.0",
|
6 |
+
author="prithiviraj damodaran",
|
7 |
+
author_email="",
|
8 |
+
description="Gramformer",
|
9 |
+
long_description="A framework for detecting, highlighting and correcting grammatical errors on natural language text",
|
10 |
+
url="https://github.com/PrithivirajDamodaran/Gramformer.git",
|
11 |
+
packages=setuptools.find_packages(),
|
12 |
+
install_requires=['transformers', 'sentencepiece==0.1.95', 'python-Levenshtein==0.12.2', 'fuzzywuzzy==0.18.0', 'tokenizers==0.10.2', 'fsspec==2021.5.0', 'lm-scorer==0.4.2', 'errant', 'st-annotated-text'],
|
13 |
+
classifiers=[
|
14 |
+
"Programming Language :: Python :: 3.7",
|
15 |
+
"License :: Apache 2.0",
|
16 |
+
"Operating System :: OS Independent",
|
17 |
+
],
|
18 |
+
)
|
19 |
+
|