Transfer learning for text classification

Deep Learning per il testo con PyTorch

Shubham Jain

Instructor

What is transfer learning?

Transfer Learning

Use pre-existing knowledge from one task to a related task

Saves time
Share expertise
Reduces need for large data

An English teacher starts teaching History

Mechanics of transfer learning

Transfer Learning I

Mechanics of transfer learning

Transfer Learning II

Mechanics of transfer learning

Transfer Learning II

Mechanics of transfer learning

Transfer Learning III

Pre-trained model : BERT

Bidirectional Encoder Representations from Transformers

Bert Sentiment Analysis

Trained for language modeling
Multiple layers of transformers
Pre-trained on large texts

Hands-on: implementing BERT

texts = ["I love this!", 
         "This is terrible.", 
         "Amazing experience!", 
         "Not my cup of tea."]
labels = [1, 0, 1, 0]

import torch
from transformers import BertTokenizer, BertForSequenceClassification

tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertForSequenceClassification.from_pretrained('bert-base-uncased',                 
                                                       num_labels=2)

inputs = tokenizer(texts, padding=True, truncation=True, 
                    return_tensors="pt", max_length=32)
inputs["labels"] = torch.tensor(labels)

Fine-tuning BERT

optimizer = torch.optim.AdamW(model.parameters(), lr=0.00001)
model.train()

for epoch in range(1):
    outputs = model(**inputs)

    loss = outputs.loss
    loss.backward()

    optimizer.step()
    optimizer.zero_grad()

    print(f"Epoch: {epoch+1}, Loss: {loss.item()}")

Epoch: 1, Loss: 0.7061821222305298

Evaluating on new text

text = "I had an awesome day!"
input_eval = tokenizer(text, return_tensors="pt", truncation=True, 
                       padding=True, max_length=128)

outputs_eval = model(**input_eval)

predictions = torch.nn.functional.softmax(outputs_eval.logits, dim=-1)

predicted_label = 'positive' if torch.argmax(predictions) > 0 else 'negative'
print(f"Text: {text}\nSentiment: {predicted_label}")

Text: I had an awesome day!
Sentiment: positive

Let's practice!

Deep Learning per il testo con PyTorch