Adding multi class classification and tuning for dense and dropout la…

neo_dolfin/ai/sentiment_analysis/BERT_multiclass.py

@@ -0,0 +1,149 @@
+# -*- coding: utf-8 -*-
+"""Untitled6.ipynb
+
+Automatically generated by Colab.
+
+Original file is located at
+    https://colab.research.google.com/drive/1mRG9RhnH8GNbDah4xHBCRAVAplv__7nR
+"""
+
+import torch
+from transformers import BertTokenizerFast, BertForSequenceClassification
+from sklearn.model_selection import train_test_split
+from torch.utils.data import Dataset, DataLoader
+from transformers import AdamW
+import matplotlib.pyplot as plt
+import pandas as pd
+
+# Load the BERT tokenizer and model
+tokenizer = BertTokenizerFast.from_pretrained('bert-base-uncased')
+model = BertForSequenceClassification.from_pretrained('bert-base-uncased', num_labels=3)
+
+# Prepare the dataset
+df = pd.read_csv('data.csv')
+df['Sentiment'] = df['Sentiment'].map({'positive': 2, 'negative': 0, 'neutral': 1})
+train_texts, test_texts, train_labels, test_labels = train_test_split(df['Sentence'], df['Sentiment'], test_size=0.2)
+
+# Tokenize the dataset
+train_encodings = tokenizer(list(train_texts), truncation=True, padding=True)
+test_encodings = tokenizer(list(test_texts), truncation=True, padding=True)
+
+# Create a PyTorch Dataset
+class SentimentDataset(Dataset):
+    def __init__(self, encodings, labels):
+        self.encodings = encodings
+        self.labels = labels
+
+    def __getitem__(self, idx):
+        item = {key: torch.tensor(val[idx]) for key, val in self.encodings.items()}
+        item['labels'] = torch.tensor(self.labels[idx])
+        return item
+
+    def __len__(self):
+        return len(self.labels)
+
+train_dataset = SentimentDataset(train_encodings, list(train_labels))
+test_dataset = SentimentDataset(test_encodings, list(test_labels))
+
+# Create a DataLoader
+train_loader = DataLoader(train_dataset, batch_size=16, shuffle=True)
+test_loader = DataLoader(test_dataset, batch_size=16, shuffle=True)
+
+# Set up the optimizer
+optimizer = AdamW(model.parameters(), lr=1e-5)
+
+# Variables to keep track of losses and accuracies for each epoch
+train_losses = []
+test_accuracies = []
+
+# Training loop
+device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
+model.to(device)
+for epoch in range(3):
+    total_loss = 0
+    model.train()
+    for batch in train_loader:
+        optimizer.zero_grad()
+        input_ids = batch['input_ids'].to(device)
+        attention_mask = batch['attention_mask'].to(device)
+        labels = batch['labels'].to(device)
+        outputs = model(input_ids, attention_mask=attention_mask, labels=labels)
+        loss = outputs[0]
+        loss.backward()
+        optimizer.step()
+        total_loss += loss.item()
+    avg_train_loss = total_loss / len(train_loader)
+    train_losses.append(avg_train_loss)
+
+    # Evaluation for each epoch
+    model.eval()
+    total_predictions = []
+    total_actual = []
+    for batch in test_loader:
+        input_ids = batch['input_ids'].to(device)
+        attention_mask = batch['attention_mask'].to(device)
+        labels = batch['labels'].to(device)
+        with torch.no_grad():
+            outputs = model(input_ids, attention_mask=attention_mask)
+        logits = outputs[0]
+        predictions = torch.argmax(logits, dim=-1)
+        total_predictions.extend(predictions.cpu().numpy())
+        total_actual.extend(labels.cpu().numpy())
+    accuracy = np.sum(np.array(total_predictions) == np.array(total_actual)) / len(total_actual)
+    test_accuracies.append(accuracy)
+
+    print(f"Epoch: {epoch}, Loss: {avg_train_loss}, Accuracy: {accuracy}")
+
+# Plot training loss over time
+plt.figure(figsize=(10, 5))
+plt.subplot(1, 2, 1)
+plt.plot(train_losses, label='Training Loss')
+plt.xlabel('Epoch')
+plt.ylabel('Loss')
+plt.title('Training Loss Over Time')
+plt.legend()
+
+# Plot test accuracy over time
+plt.subplot(1, 2, 2)
+plt.plot(test_accuracies, label='Test Accuracy')
+plt.xlabel('Epoch')
+plt.ylabel('Accuracy')
+plt.title('Test Accuracy Over Time')
+plt.legend()
+
+plt.tight_layout()
+plt.show()
+
+# Save the model
+torch.save(model.state_dict(), 'best_model_1.pt')
+
+# Calculate sentiment counts in training and testing data
+train_counts = train_labels.value_counts()
+test_counts = test_labels.value_counts()
+
+# Create subplots
+fig, axs = plt.subplots(1, 2, figsize=(12, 6))
+
+# Plot training data distribution
+axs[0].bar(train_counts.index, train_counts.values)
+axs[0].set_xticks([0, 1, 2])
+axs[0].set_xticklabels(['negative', 'neutral', 'positive'])
+axs[0].set_title('Training Data Distribution')
+axs[0].set_xlabel('Sentiment')
+axs[0].set_ylabel('Count')
+
+# Plot testing data distribution
+axs[1].bar(test_counts.index, test_counts.values)
+axs[1].set_xticks([0, 1, 2])
+axs[1].set_xticklabels(['negative', 'neutral', 'positive'])
+axs[1].set_title('Testing Data Distribution')
+axs[1].set_xlabel('Sentiment')
+axs[1].set_ylabel('Count')
+
+# Show the plots
+plt.tight_layout()
+plt.show()
+
+"""# References
+https://medium.com/nerd-for-tech/fine-tuning-pretrained-bert-for-sentiment-classification-using-transformers-in-python-931ed142e37
+"""