BERT モデルを感情分類に使用する方法は何ですか?

!wget http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz
!tar -xf aclImdb_v1.tar.gz

import torch
import transformers as ppb
import numpy as np

<code>model_class, tokenizer_class, pretrained_weights = (ppb.BertModel, ppb.BertTokenizer, &#x27;bert-base-uncased&#x27;)<br/>tokenizer = tokenizer_class.from_pretrained(pretrained_weights)<br/>model = model_class.from_pretrained(pretrained_weights)</code>

import pandas as pd
import io

# Load data
train = pd.read_csv('aclImdb/train.tsv', delimiter='\t', header=None)
test = pd.read_csv('aclImdb/test.tsv', delimiter='\t', header=None)

# Split data into input and labels
train_sentences = train[0].values
train_labels = train[1].values
test_sentences = test[0].values
test_labels = test[1].values

# Tokenize the input texts
train_tokenized = np.array([tokenizer.encode(sent, add_special_tokens=True) for sent in train_sentences])
test_tokenized = np.array([tokenizer.encode(sent, add_special_tokens=True) for sent in test_sentences])

# Truncate and pad the input texts
max_len = 128
train_padded = np.array([i[:max_len] + [0]*(max_len-len(i)) for i in train_tokenized])
test_padded = np.array([i[:max_len] + [0]*(max_len-len(i)) for i in test_tokenized])

# Create attention masks
train_attention_mask = np.where(train_padded != 0, 1, 0)
test_attention_mask = np.where(test_padded != 0, 1, 0)

# Convert the input texts to PyTorch tensors
train_input_ids = torch.tensor(train_padded)
train_attention_mask = torch.tensor(train_attention_mask)
train_labels = torch.tensor(train_labels)
test_input_ids = torch.tensor(test_padded)
test_attention_mask = torch.tensor(test_attention_mask)
test_labels = torch.tensor(test_labels)

from torch.utils.data import TensorDataset, DataLoader, RandomSampler, SequentialSampler
from transformers import AdamW, get_linear_schedule_with_warmup

#Create a data loader for training data
batch_size = 32
train_data = TensorDataset(train_input_ids, train_attention_mask, train_labels)
train_sampler = RandomSampler(train_data)
train_dataloader = DataLoader(train_data, sampler=train_sampler, batch_size=batch_size)

#Create a data loader for test data
test_data = TensorDataset(test_input_ids, test_attention_mask, test_labels)
test_sampler = SequentialSampler(test_data)
test_dataloader = DataLoader(test_data, sampler=test_sampler, batch_size=batch_size)

#Set up the optimizer and scheduler
epochs = 3
optimizer = AdamW(model.parameters(), lr=2e-5, eps=1e-8)
total_steps = len(train_dataloader) * epochs
scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=0, num_training_steps=total_steps)

#Train the model
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model.to(device)
for epoch in range(epochs):
    print(f'Epoch {epoch + 1}/{epochs}')
    print('-' * 10)
    total_loss = 0
    model.train()
    for step, batch in enumerate(train_dataloader):
        # Get batch input data
        batch_input_ids = batch[0].to(device)
        batch_attention_mask = batch[1].to(device)
        batch_labels = batch[2].to(device)

    # Clear gradients
    model.zero_grad()

    # Forward pass
    outputs = model(batch_input_ids, attention_mask=batch_attention_mask, labels=batch_labels)
    loss = outputs[0]

    # Backward pass
    loss.backward()

    # Update parameters
    optimizer.step()

    # Update learning rate schedule
    scheduler.step()

    # Accumulate total loss
    total_loss += loss.item()

    # Print progress every 100 steps
    if (step + 1) % 100 == 0:
        print(f'Step {step + 1}/{len(train_dataloader)}: Loss = {total_loss / (step + 1):.4f}')

# Evaluate the model on test data
model.eval()
with torch.no_grad():
    total_correct = 0
    total_samples = 0
    for batch in test_dataloader:
        # Get batch input data
        batch_input_ids = batch[0].to(device)
        batch_attention_mask = batch[1].to(device)
        batch_labels = batch[2].to(device)

        # Forward pass
        outputs = model(batch_input_ids, attention_mask=batch_attention_mask)
        logits = outputs[0]
        predictions = torch.argmax(logits, dim=1)

        # Accumulate total correct predictions and samples
        total_correct += torch.sum(predictions == batch_labels).item()
        total_samples += len(batch_labels)

    # Print evaluation results
    accuracy = total_correct / total_samples
    print(f'Test accuracy: {accuracy:.4f}')