-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
56 lines (52 loc) · 1.61 KB
/
utils.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
import torch
def load_data_tokenize(path):
'''
A script for loading from a corpora from a text file and tokenize it. It adds automatically the end-of-sentence special token <eos>
at the end of the sentence.
'''
tokens = []
with open(path, 'r') as f:
corpus = f.readlines()
for line in corpus:
tokens.extend(line.split() + ['<eos>'])
return tokens
def save_model(model, path):
'''
Script to save the model given the path where to save it
'''
torch.save({
"state_dict": model.state_dict()
}, path)
def detach_hidden(hidden):
'''
Script to detach the hidden state to prevent the backward step to reach the beginning of the text
'''
if len(hidden) > 2:
detached = []
for h in hidden:
detached.append(tuple([h[0].detach(), h[1].detach()]))
else:
detached = tuple([hidden[0].detach(), hidden[1].detach()])
return detached
def concat_name(model, asgd, clip_gradient, tye_weights, dropout, dropout_emb, dropout_wgt, dropout_inp, dropout_hid):
'''
Utility script to generate the name of the model weight given the regularizations used
'''
name = model
if asgd:
name += "_asgd"
if clip_gradient:
name += "_clipgradient"
if tye_weights:
name += "_tyeweights"
if dropout != 0.0:
name += "_dropout"
if dropout_emb != 0.0:
name += "_dropoutemb"
if dropout_hid != 0.0:
name += "_dropouthid"
if dropout_wgt != 0.0:
name += "_dropoutwgt"
if dropout_inp != 0.0:
name += "_dropoutinp"
return name