-
Notifications
You must be signed in to change notification settings - Fork 0
/
new_preprocess.py
275 lines (226 loc) · 7.19 KB
/
new_preprocess.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
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
import json
import math
import os
import random
import re
import unicodedata
from collections import Counter
from itertools import chain, zip_longest
from multiprocessing import Pool
import numpy as np
import pandas as pd
import spacy
from spacy.tokenizer import Tokenizer
from tqdm import tqdm
from data_viz import dataset_stats
from utils import flatten_1_deg, loadpkl, savepkl
# file paths
ALL_TABLES_PATH_ORG = '../global_data/tables_redi2_1/'
OUTPUT_DIR = '../global_data/all_tables'
all_tables = os.listdir(OUTPUT_DIR)
MAX_COL_LEN = 10
MAX_ROW_LEN = 50
SAMPLE_PERC = 0.5
LENGTH_PER_CELL = 20
nlp = spacy.load("en_core_web_md")
nlp.add_pipe(nlp.create_pipe("merge_entities"))
nlp.add_pipe(nlp.create_pipe("merge_noun_chunks"))
def read_table(table):
if table.split('.')[-1] == 'json':
table = table.split('.')[0]
with open(os.path.join(OUTPUT_DIR, f"{table}.json"), 'r') as f:
j = json.load(f)
return j
def clean_entities(inp):
if len(inp):
if inp[0] == '[' and inp[-1] == ']':
inp = inp.split('|')[0][1:] # Take the 1st element
# inp = inp.split('|')[-1][:-1] # Take the 2nd element
return inp
else:
return inp
def split_data(data):
data = np.array(data)
(row_shape, column_shape) = data.shape
blocks_per_row = math.ceil(row_shape/MAX_ROW_LEN)
blocks_per_column = math.ceil(column_shape/MAX_COL_LEN)
previous_row = 0
for row_block in range(blocks_per_row):
previous_row = row_block * MAX_ROW_LEN
previous_column = 0
for column_block in range(blocks_per_column):
previous_column = column_block * MAX_COL_LEN
block = data[previous_row:previous_row + MAX_ROW_LEN,
previous_column:previous_column + MAX_COL_LEN]
yield block
def split_overflow_table(j):
X = []
if j['numCols'] == 0 or j['numDataRows'] == 0:
# print('Empty tables')
return X
if j['numCols'] > MAX_COL_LEN or j['numDataRows'] > MAX_ROW_LEN:
# print('Splitting the data')
splits = split_data(j['data'])
for v in splits:
if v.shape[0] != 0 or v.shape[1] != 0:
# print('Adding split data')
X.append(v.tolist())
else:
X.append(j['data'])
return X
def tokenize_table(table):
for i, row in enumerate(table):
for j, cell in enumerate(row):
table[i][j] = tokenize_str(clean_entities(cell))
# table = filter_empty_cols(table)
return table
def tokenize_str(cell):
a = unicodedata.normalize('NFKD', cell).encode(
'ascii', 'ignore').decode('utf-8')
t = [token.orth_ for token in nlp(a) if not (
token.is_punct
or len(token.orth_) < 4
or token.is_space
or token.is_stop
or token.is_currency
or token.like_url
or token.like_email
or token.like_num
or small_alphanum(token.orth_)
or token.ent_type_ in ['DATE', 'TIME', 'MONEY', 'PERCENT']
)]
t = [i.replace(" ", "_") for i in t]
return t
def contains_num(s):
return any(c.isdigit() for c in s)
def small_alphanum(s):
if len([i for i in s if i.isalpha()]) < 3:
return True
else:
return False
def cell_overflow_cap(X):
def clip(cell):
if len(cell) > LENGTH_PER_CELL:
return cell[:LENGTH_PER_CELL]
else:
return cell
for t, table in enumerate(X):
for i, row in enumerate(table):
for j, cell in enumerate(row):
table[i][j] = clip(cell)
return X
# def column_filter(table):
# def check_cell_validity(column):
# c = 0
# for i in column:
# if len(i) < 4:
# c += 1
# if contains_num(i):
# return True
# if c/len(column) > 0.3:
# return True
# return False
# data = np.array(table)
# cols = data.shape[1]
# col = 0
# while(col < data.shape[1]):
# if check_cell_validity(data[:, col]):
# data = np.delete(data, col, 1)
# else:
# col += 1
# return data.tolist()
def filter_empty_cols(table):
def check_cell_validity(column):
c = 0
for i in column:
if len(i) == 0:
c += 1
r = c/len(column)
if len(column) < 20 and r >= 0.33:
return True
elif len(column) >= 20 and c/len(column) >= 0.5:
return True
return False
data = np.array(table)
cols = data.shape[1]
col = 0
while(col < data.shape[1]):
if check_cell_validity(data[:, col]):
data = np.delete(data, col, 1)
else:
col += 1
return data.tolist()
def print_table(table):
for row in table:
for col in row:
print(col)
print()
def remove_empty_tables(tables):
tables = np.array(tables)
e_t = []
for i in range(len(tables)):
if np.array(tables[i]).size == 0:
e_t.append(i)
tables = np.delete(tables, e_t)
return tables.tolist()
def generate_vocab(X):
'''
Generating word distribution dataframe
'''
result = flatten_1_deg(flatten_1_deg(flatten_1_deg(X)))
query_l = [tokenize_str(i) for i in list(baseline_f['query'].unique())]
query_l = flatten_1_deg(query_l)
result += query_l
# print(result[:10])
count = Counter(result)
c = [[i, count[i]] for i in count.keys()]
df = pd.DataFrame(c)
df.sort_values(by=[1], ascending=False, inplace=True)
df.to_csv('./data/word_distr_2D_complete.csv', index=False, columns=None)
'''
Getting the vocab from the data
'''
vocab = list(set(count.keys()))
vocab.insert(0, '<UNK>')
vocab.insert(0, '<PAD>')
print(f'vocab: {len(vocab)}\n')
savepkl(
f'./data/vocab_{MAX_COL_LEN}-{MAX_ROW_LEN}.pkl', vocab)
def data_prep_pipeline(X):
'''
Breaking tables into chunks over max col and row length
'''
# X = [split_overflow_table(read_table(table)) for table in X]
# print(f"Intial # of tables before splitting::: {len(X)}")
# X = flatten_1_deg(X)
# print(f"Final # of tables::: {len(X)}")
'''
Tokenizing and filtering out empty columns from X
'''
# with Pool(50) as p:
# X = [tqdm(p.imap(tokenize_table, X), total=len(X))]
p = Pool(processes=75)
X = p.map(tokenize_table, X)
p.close()
p.join()
# print_table(X[0])
# '''
# Remove totally empty tables, generating vocab and cliiping cells to max_cell_len
# '''
# generate_vocab(X)
# X = remove_empty_tables(X)
# X = cell_overflow_cap(X)
X = np.array(X)
print(X.shape)
# print_table(X[0])
return X
if __name__ == "__main__":
baseline_f = pd.read_csv('../global_data/features.csv')
tables_subset_3k = list(baseline_f['table_id'])
tables_subset = list(
set(tables_subset_3k+random.sample(all_tables, 20000)))
savepkl(f'./data/wo_strnum3.0_wo_ent/postive_tables_set.pkl', tables_subset)
read_all_tables = [read_table(js)['data'] for js in tables_subset]
X = data_prep_pipeline(read_all_tables)
savepkl(f'./data/wo_strnum3.0_wo_ent/x_tokenised.pkl', X)
# savepkl(f'./data/xp_{MAX_COL_LEN}-{MAX_ROW_LEN}.pkl', X)