-
Notifications
You must be signed in to change notification settings - Fork 0
/
Embedding.py
53 lines (38 loc) · 1.61 KB
/
Embedding.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
from decoder import Decoder
from encoder import Encoder
from runner import Runner
from settings import LSTM_HIDDEN_SIZE, EMBEDDED_SIZE
from utils import load_datasets, get_device
from glove import loadGlove
import numpy as np
from torch import nn
import torch
def getPreTrainedEmbeddingRunner():
embed_size = 50
glove = loadGlove()
print("Loaded glove")
train_dataset, val_dataset, test_dataset = load_datasets()
vocabulary_size = len(train_dataset.dataset.vocab.wordToIndex)
glove_weights = np.zeros((vocabulary_size, embed_size))
missed_words = 0
found_words = 0
for i, word in enumerate(train_dataset.dataset.vocab.wordToIndex.keys()):
try:
glove_weights[i] = glove[word]
found_words += 1
except KeyError:
glove_weights[i] = np.random.normal(scale=0.6, size=(embed_size, ))
missed_words += 1
print("{} words not in glove".format(missed_words))
print("{} words found".format(found_words))
computing_device = get_device()
encoder = Encoder(embed_size).to(computing_device)
decoder = Decoder(embed_size, LSTM_HIDDEN_SIZE, vocabulary_size).to(computing_device)
(num_embed, embed_dim) = glove_weights.shape
embed_layer = nn.Embedding(num_embed, embed_dim).to(computing_device)
#glove_weights = torch.tensor(glove_weights).to(computing_device)
embed_layer.load_state_dict({'weight': torch.tensor(glove_weights)})
embed_layer.weight.requires_grad = False
decoder.embedding = embed_layer
runner = Runner(encoder, decoder, train_dataset, val_dataset, test_dataset)
return runner