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
| import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import Dataset, DataLoader
import pandas as pd
import numpy as np
import nltk
from nltk.tokenize import word_tokenize
from sklearn.model_selection import train_test_split
from tqdm import tqdm
class Word2VecDataset(Dataset):
def __init__(self, pairs):
self.pairs = pairs
def __len__(self):
return len(self.pairs)
def __getitem__(self, idx):
return self.pairs[idx]
class Word2Vec(nn.Module):
def __init__(self, vocab_size, embedding_dim):
super().__init__()
self.in_embed = nn.Embedding(vocab_size, embedding_dim)
self.out_embed = nn.Embedding(vocab_size, embedding_dim)
def forward(self, center_idx):
center_vec = self.in_embed(center_idx)
scores = torch.matmul(center_vec, self.out_embed.weight.T)
log_probs = nn.functional.log_softmax(scores, dim=1)
return log_probs
class SentimentDataset(Dataset):
def __init__(self, phrases, labels, word_to_vec, dim):
self.vectors = []
self.labels = labels.values if labels is not None else None
for phrase in tqdm(phrases, desc="Processing phrases for Dataset", ncols=80):
tokens = word_tokenize(phrase.lower())
vectors = [word_to_vec[word] for word in tokens if word in word_to_vec]
if vectors:
avg_vec = np.mean(vectors, axis=0)
else:
avg_vec = np.zeros(dim)
self.vectors.append(torch.tensor(avg_vec, dtype=torch.float32))
def __len__(self):
return len(self.vectors)
def __getitem__(self, idx):
if self.labels is not None:
return self.vectors[idx], self.labels[idx]
else:
return self.vectors[idx]
class SentimentClassifier(nn.Module):
def __init__(self, input_dim, output_dim):
super().__init__()
self.fc = nn.Linear(input_dim, output_dim)
def forward(self, x):
return self.fc(x)
if __name__ == '__main__':
train_df = pd.read_csv("train.tsv", sep="\t")
test_df = pd.read_csv("test.tsv", sep="\t")
all_phrases = pd.concat([train_df['Phrase'], test_df['Phrase']], ignore_index=True)
tokenized_phrases = [word_tokenize(p.lower()) for p in all_phrases]
tokens = [word for phrase in tokenized_phrases for word in phrase]
vocab = sorted(set(tokens))
word_to_idx = {word: i for i, word in enumerate(vocab)}
vocab_size = len(vocab)
window_size = 2
training_pairs = []
for phrase in tqdm(tokenized_phrases, desc="Building training pairs", ncols=80):
for i, center_word in enumerate(phrase):
if center_word not in word_to_idx:
continue
center_idx = word_to_idx[center_word]
for j in range(-window_size, window_size + 1):
if j == 0 or i + j < 0 or i + j >= len(phrase):
continue
context_word = phrase[i + j]
if context_word in word_to_idx:
context_idx = word_to_idx[context_word]
training_pairs.append((center_idx, context_idx))
embedding_dim = 100
w2v_model = Word2Vec(vocab_size, embedding_dim)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"using {device}")
w2v_model.to(device)
criterion = nn.NLLLoss()
optimizer = optim.Adam(w2v_model.parameters(), lr=0.001)
batch_size = 512
train_dataset = Word2VecDataset(training_pairs)
train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True, num_workers=2)
for epoch in range(5):
w2v_model.train()
total_loss = 0
for center_batch, context_batch in tqdm(train_loader, desc=f"Word2Vec Epoch {epoch+1}", ncols=80):
center_batch = center_batch.to(device)
context_batch = context_batch.to(device)
optimizer.zero_grad()
log_probs = w2v_model(center_batch)
loss = criterion(log_probs, context_batch)
loss.backward()
optimizer.step()
total_loss += loss.item() * center_batch.size(0)
avg_loss = total_loss / len(train_dataset)
print(f"Epoch {epoch+1}, Loss: {avg_loss:.4f}")
word_vectors = w2v_model.in_embed.weight.data.cpu().numpy()
word_to_vec = {word: word_vectors[word_to_idx[word]] for word in vocab}
train_data, val_data = train_test_split(train_df, test_size=0.1, random_state=42)
train_dataset = SentimentDataset(train_data['Phrase'], train_data['Sentiment'], word_to_vec, embedding_dim)
val_dataset = SentimentDataset(val_data['Phrase'], val_data['Sentiment'], word_to_vec, embedding_dim)
test_dataset = SentimentDataset(test_df['Phrase'], None, word_to_vec, embedding_dim)
train_loader = DataLoader(train_dataset, batch_size=64, shuffle=True)
val_loader = DataLoader(val_dataset, batch_size=64)
test_loader = DataLoader(test_dataset, batch_size=64)
model = SentimentClassifier(embedding_dim, 5).to(device)
loss_fn = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.001)
for epoch in range(10):
model.train()
total_loss = 0
for x, y in tqdm(train_loader, desc=f"Sentiment Training Epoch {epoch+1}", ncols=80):
x, y = x.to(device), torch.tensor(y).to(device)
optimizer.zero_grad()
out = model(x)
loss = loss_fn(out, y)
loss.backward()
optimizer.step()
total_loss += loss.item()
print(f"Epoch {epoch+1}, Loss: {total_loss / len(train_loader):.4f}")
model.eval()
all_preds = []
with torch.no_grad():
for x in tqdm(test_loader, desc="Predicting Test Set", ncols=80):
x = x.to(device)
out = model(x)
preds = torch.argmax(out, dim=1)
all_preds.extend(preds.cpu().numpy())
submission = pd.DataFrame({
'PhraseId': test_df['PhraseId'],
'Sentiment': all_preds
})
submission.to_csv("sampleSubmission.csv", index=False)
print("预测完成，结果已保存为 sampleSubmission.csv")
|
