-
Notifications
You must be signed in to change notification settings - Fork 11
/
Pipeline.py
127 lines (118 loc) · 6.79 KB
/
Pipeline.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
from argparse import Namespace
import torch
from utils import get_device, base_context, split_sentence, collate_fn_base, collate_fn_demo, TEXT_COLOR, BINARY_COLOR, UNDER_COLOR
from .demo_utils import transform2inform, context2docs
import time
from .ReportDataset import ReportDataset
from .ModelBucket import ModelBucketV1
from torch.utils.data import DataLoader
import numpy as np
from docxtpl import RichText
class Pipeline(object):
def __init__(self, args_binary : Namespace, args_demo : Namespace):
self.args_binary = args_binary
self.args_demo = args_demo
# self.device = get_device(args_demo.cuda, args_demo.gpu_id)
self.device = torch.device("cpu") # RECOMMAND
self.model_bucket = ModelBucketV1(args_demo, self.device, binary=True, switch=True, taggen=True, checkpoints_name='checkpoint.pt')
print(">> Model initialized.")
def __call__(self, documents : str, report_name : str = None, export_path : str = None, doc_type : str= 'pdf') -> dict:
'''
Pipline processor for fcgec reporter
:param documents: input docs [str]
:return: path [str]
'''
context = base_context(report_name, documents, switch=True, modify=True)
sentences = split_sentence(documents, self.args_demo.padding_size)
data_bucket = ReportDataset(self.args_demo, sentences)
# Binary Judge
start_time = time.time()
data_bucket.binary()
# binary_results, types_results = self.binary_report(data_bucket)
switch_results, switch_tags, switch_pointers = self.switch_report(data_bucket)
switch_inform = (switch_results, switch_tags, switch_pointers)
# Tagger Judge# Generate fake labels
data_bucket.tagger(switch_results, switch_tags)
tagger_results, generate_tags = self.tagger_process(data_bucket)
# Generate Judge
data_bucket.generate(generate_tags)
generate_results, modified_text = self.generate_process(data_bucket)
# Details Generator
details_modification = transform2inform(data_bucket, generate_results, (tagger_results, generate_tags), switch_inform)
detail_texts = self.details_generate(sentences, operate_reports=details_modification)
context['details'] = self.pack_textwtype(sentences, detail_texts)
context['process_time'] = '{:.5f}s'.format(time.time() - start_time)
# Modified Sentence
context['modify_text'] = self.process_modified_sentence(data_bucket, modified_text, generate_tags[0])
path = context2docs(context, export_path, doc_type)
return path
def binary_report(self, data_bucket : ReportDataset) -> tuple:
BinaryLoader = DataLoader(data_bucket, batch_size=self.args_binary.batch_size, shuffle=False, drop_last=False, collate_fn=collate_fn_base)
results, type_collect = [], []
for _, batch_data in enumerate(BinaryLoader):
tokens = batch_data
result, types = self.model_bucket.binary_process(tokens)
results.extend(result)
type_collect.extend(types)
return results, type_collect
def switch_report(self, data_bucket : ReportDataset) -> tuple:
SwitchLoader = DataLoader(data_bucket, batch_size=self.args_demo.batch_size, shuffle=False, drop_last=False,collate_fn=collate_fn_demo)
results, sw_flag_collect, sw_preds_collection = [], [], []
for _, batch_data in enumerate(SwitchLoader):
tokens = batch_data
result, sw_preds = self.model_bucket.switch_process(tokens)
results.extend(result)
sw_preds_collection.extend(sw_preds)
sw_flag_collect.extend([True if np.max(np.diff(sw_preds[idx])) > 1 else False for idx in range(len(batch_data))])
return results, sw_flag_collect, sw_preds_collection
def tagger_process(self, data_bucket : ReportDataset) -> tuple:
TaggerLoader = DataLoader(data_bucket, batch_size=self.args_demo.batch_size, shuffle=False, drop_last=False, collate_fn=collate_fn_demo)
results, gen_inform = [[], []], [[], []]
for _, batch_data in enumerate(TaggerLoader):
tokens = batch_data
result, gen = self.model_bucket.tagger_process(tokens)
results[0].extend(result[0])
results[1].extend(result[1])
gen_inform[0].extend(gen[0])
gen_inform[1].extend(gen[1])
return results, gen_inform
def generate_process(self, data_bucket : ReportDataset) -> tuple:
GenerateLoader = DataLoader(data_bucket, batch_size=self.args_demo.batch_size, shuffle=False, drop_last=False, collate_fn=collate_fn_demo)
results_token, modified_sentences = [], []
for _, batch_data in enumerate(GenerateLoader):
result, modified = self.model_bucket.generate_process(batch_data)
results_token.extend(result)
modified_sentences.extend(modified)
return results_token, modified_sentences
def details_generate(sself, sentences : list, binary_reports : list = None, operate_reports : list = None) -> list:
detail_text, global_idx = [], 1
assert operate_reports is not None
for index in range(len(sentences)):
rt = RichText()
if operate_reports[index][0] == '该句子没有语病': rt.add(operate_reports[index][0])
else:
op_str = ''
for opidx in range(len(operate_reports[index])):op_str += ('[{}] {}\n'.format(global_idx, operate_reports[index][opidx]))
rt.add(op_str, color=TEXT_COLOR)
global_idx += 1
detail_text.append(rt)
return detail_text
def pack_textwtype(self, detail_text : list, detail_type : list) -> list:
pack_details = []
assert len(detail_text) == len(detail_type)
detail_num = len(detail_text)
for didx in range(detail_num):
rt_text = RichText()
texts, types = detail_text[didx], detail_type[didx]
rt_text.add(texts, color=TEXT_COLOR)
pack_details.append({'detail_text' : rt_text, 'detail_type' : types})
return pack_details
def process_modified_sentence(self, data_bucket: ReportDataset, modified_sentences: list, tagger_sentences: list):
sentences, genidx = [], 0
for idx in range(len(data_bucket.filter_flag)):
flag = data_bucket.filter_flag[idx]
if flag: sentences.append(''.join(data_bucket.tokenizer.convert_ids_to_tokens(tagger_sentences[idx][1:-1])).replace('##', '').replace('[UNK]', '"').replace('[PAD]', '').replace('[SEP]', '"'))
else:
sentences.append(''.join(data_bucket.tokenizer.convert_ids_to_tokens(modified_sentences[genidx][1:-1])).replace('##', '').replace('[UNK]', '"').replace('[PAD]', '').replace('[SEP]', '"'))
genidx += 1
return ''.join(sentences)