forked from naiveHobo/InvoiceNet
-
Notifications
You must be signed in to change notification settings - Fork 1
/
data_handler.py
125 lines (106 loc) · 4.61 KB
/
data_handler.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
import numpy as np
import pickle
import gzip
from gensim.models import Word2Vec
class DataHandler:
def __init__(self, data=None, max_len=10):
self.data = data
self.max_length = max_len
self.vocab_size = 0
self.word2idx = {}
self.idx2word = {}
self.embeddings = None
self.embed_size = 300
self.PAD = '<pad>'
self.UNKNOWN = '<unk>'
self.START = '<start>'
self.END = '<end>'
self.label_dict = {0: 0, 1: 1, 2: 2, 8: 3, 14: 4, 18: 5}
self.num_classes = len(self.label_dict)
self.train_data = {}
# self.type_dict = {'text': 0.1, 'number': 0.2, 'email': 0.3, 'date': 0.4, '': 0.5, 'money': 0.6, 'phone': 0.7}
def read(self, data, max_len=10):
"""Read DataFrame"""
self.data = data
self.max_length = max_len
def process_data(self, tokens, coordinates):
tokens = [self.START] + tokens[:self.max_length - 2] + [self.END]
tokens += [self.PAD] * (self.max_length - len(tokens))
inp = np.array([self.get_word_id(token) for token in tokens])
coordinates = np.array(coordinates)
return inp, coordinates
def prepare_data(self):
"""Prepares data for training"""
inputs = []
labels = []
coordinates = []
for i, row in self.data.iterrows():
text = row['type']
coords = row['coords']
label = self.label_dict[int(row['label'])]
tokens = text[0].strip().split(' ')
# dtypes = [self.type_dict[dtype] for dtype in text[1].split(',')]
height = float(text[-2])
width = float(text[-1])
min_x = float(coords[0]) / width
min_y = float(coords[1]) / height
max_x = float(coords[2]) / width
max_y = float(coords[3]) / height
tokens = [self.START] + tokens[:self.max_length - 2] + [self.END]
tokens += [self.PAD] * (self.max_length - len(tokens))
inp = [self.get_word_id(token) for token in tokens]
inputs.append(np.array(inp))
labels.append(np.array(label))
coordinates.append(np.array([min_x, min_y, max_x, max_y]))
self.train_data['inputs'] = np.array(inputs)
self.train_data['labels'] = np.array(labels)
self.train_data['coordinates'] = np.array(coordinates)
def load_embeddings(self, model_path):
"""Loads pre-trained gensim model"""
print("\nLoading pre-trained embeddings...")
model = Word2Vec.load(model_path)
words = list(model.wv.vocab)
embed_size = model.layer1_size
embed = []
word2idx = {self.PAD: 0, self.UNKNOWN: 1, self.START: 2, self.END: 3}
idx2word = {0: self.PAD, 1: self.UNKNOWN, 2: self.START, 3: self.END}
embed.append(np.zeros(embed_size, dtype=np.float32))
embed.append(np.random.uniform(-0.1, 0.1, embed_size))
embed.append(np.random.uniform(-0.1, 0.1, embed_size))
embed.append(np.random.uniform(-0.1, 0.1, embed_size))
for word in words:
vector = model.wv[word]
embed.append(vector)
word2idx[word] = len(word2idx)
idx2word[word2idx[word]] = word
self.vocab_size = len(word2idx)
self.word2idx = word2idx
self.idx2word = idx2word
self.embeddings = np.array(embed, dtype=np.float32)
print("\nSuccessfully loaded pre-trained embeddings!")
def get_word_id(self, token):
"""Returns the id of a token"""
token = token.lower()
if token in self.word2idx:
return self.word2idx[token]
return self.word2idx[self.UNKNOWN]
def save_data(self, out_path='./data/processed.pkl.gz'):
"""Saves the embeddings and vocab as a zipped pickle file"""
assert (self.embeddings is not None or self.word2idx), "Data has not been processed yet"
pkl = {'embeddings': self.embeddings,
'word2idx': self.word2idx,
'idx2word': self.idx2word
}
with gzip.open(out_path, 'wb') as out_file:
pickle.dump(pkl, out_file)
print("\nData stored as {}".format(out_path))
def load_data(self, path):
"""Loads embeddings and vocab from a zipped pickle file"""
with gzip.open(path, 'rb') as in_file:
pkl = pickle.load(in_file)
self.embeddings = pkl['embeddings']
self.embed_size = self.embeddings.shape[1]
self.word2idx = pkl['word2idx']
self.vocab_size = len(self.word2idx)
self.idx2word = pkl['idx2word']
print("\nSuccessfully loaded data from {}".format(path))