-
Notifications
You must be signed in to change notification settings - Fork 0
/
part3.py
138 lines (105 loc) · 4.66 KB
/
part3.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
import numpy as np
import torch
import torch.nn as tnn
import torch.nn.functional as F
import torch.optim as topti
from torchtext import data
from torchtext.vocab import GloVe
from imdb_dataloader import IMDB
# Class for creating the neural network.
class Network(tnn.Module):
"""
Conv -> ReLu -> maxpool(size=4) -> Conv -> ReLu -> maxpool(size=4) ->
Conv -> ReLu -> maxpool over time (global pooling) -> Linear(1)
"""
def __init__(self):
super(Network, self).__init__()
self.lstm = torch.nn.LSTM(50, 100, batch_first=True)
self.lstm1 = torch.nn.LSTM(100, 200, batch_first=True)
self.ln1 = torch.nn.Linear(200, 100)
self.ln2 = torch.nn.Linear(100, 64)
self.ln3 = torch.nn.Linear(64, 1)
def forward(self, input, length):
"""
DO NOT MODIFY FUNCTION SIGNATURE
Create the forward pass through the network.
"""
sent_packed = tnn.utils.rnn.pack_padded_sequence(input=input, lengths=length, batch_first=True)
output, (h_0, c_0) = self.lstm(sent_packed)
output, (h_0, c_0) = self.lstm1(output)
x = h_0
x = tnn.functional.relu(self.ln1(x))
x = tnn.functional.relu(self.ln2(x))
x = self.ln3(x)
x = x.view(-1)
return x
class PreProcessing():
def pre(x):
"""Called after tokenization"""
return x
def post(batch, vocab):
"""Called after numericalization but prior to vectorization"""
return batch
text_field = data.Field(lower=True, include_lengths=True, batch_first=True, preprocessing=pre, postprocessing=post)
def lossFunc():
"""
Define a loss function appropriate for the above networks that will
add a sigmoid to the output and calculate the binary cross-entropy.
"""
return tnn.BCEWithLogitsLoss()
def main():
# Use a GPU if available, as it should be faster.
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print("Using device: " + str(device))
# Load the training dataset, and create a data loader to generate a batch.
textField = PreProcessing.text_field
labelField = data.Field(sequential=False)
train, dev = IMDB.splits(textField, labelField, train="train", validation="dev")
textField.build_vocab(train, dev, vectors=GloVe(name="6B", dim=50))
labelField.build_vocab(train, dev)
trainLoader, testLoader = data.BucketIterator.splits((train, dev), shuffle=True, batch_size=64,
sort_key=lambda x: len(x.text), sort_within_batch=True)
net = Network().to(device)
criterion =lossFunc()
optimiser = topti.Adam(net.parameters(), lr=0.001) # Minimise the loss using the Adam algorithm.
for epoch in range(10):
running_loss = 0
for i, batch in enumerate(trainLoader):
# Get a batch and potentially send it to GPU memory.
inputs, length, labels = textField.vocab.vectors[batch.text[0]].to(device), batch.text[1].to(
device), batch.label.type(torch.FloatTensor).to(device)
labels -= 1
# PyTorch calculates gradients by accumulating contributions to them (useful for
# RNNs). Hence we must manually set them to zero before calculating them.
optimiser.zero_grad()
# Forward pass through the network.
output = net(inputs, length)
loss = criterion(output, labels)
# Calculate gradients.
loss.backward()
# Minimise the loss according to the gradient.
optimiser.step()
running_loss += loss.item()
if i % 32 == 31:
print("Epoch: %2d, Batch: %4d, Loss: %.3f" % (epoch + 1, i + 1, running_loss / 32))
running_loss = 0
num_correct = 0
# Save mode
torch.save(net.state_dict(), "./model.pth")
print("Saved model")
# Evaluate network on the test dataset. We aren't calculating gradients, so disable autograd to speed up
# computations and reduce memory usage.
with torch.no_grad():
for batch in testLoader:
# Get a batch and potentially send it to GPU memory.
inputs, length, labels = textField.vocab.vectors[batch.text[0]].to(device), batch.text[1].to(
device), batch.label.type(torch.FloatTensor).to(device)
labels -= 1
# Get predictions
outputs = torch.sigmoid(net(inputs, length))
predicted = torch.round(outputs)
num_correct += torch.sum(labels == predicted).item()
accuracy = 100 * num_correct / len(dev)
print(f"Classification accuracy: {accuracy}")
if __name__ == '__main__':
main()