Skip to content

README.md

Latest commit

 

History

History
443 lines (349 loc) · 13.5 KB

README.md

File metadata and controls

443 lines (349 loc) · 13.5 KB

Examples

This directory contains comprehensive examples demonstrating how to use the torchTextClassifiers package for various text classification tasks.

📁 Available Examples

1. Basic Classification

A simple binary sentiment classification example that covers:

  • Creating a FastText classifier
  • Preparing training and validation data
  • Building and training the model
  • Making predictions and evaluating performance
  • Saving model configuration

Run the example:

cd examples
uv run --extra huggingface python basic_classification.py

What you'll learn:

  • Basic API usage
  • Binary classification workflow
  • Model evaluation
  • Configuration persistence

2. Multi-class Classification

Demonstrates 3-class sentiment analysis (positive, negative, neutral):

  • Multi-class data preparation
  • Class distribution handling
  • Detailed result analysis
  • Configuration loading and validation

Run the example:

cd examples
uv run  --extra huggingface python multiclass_classification.py

What you'll learn:

  • Multi-class classification setup
  • Class imbalance considerations
  • Advanced result interpretation
  • Model serialization/deserialization

3. Using Additional Features

Shows how to combine text and categorical features:

  • Text + categorical data preparation
  • Feature engineering for categorical variables
  • Comparing mixed vs. text-only models
  • Performance analysis with different feature types

Run the example:

cd examples
uv run  --extra huggingface python Using_additional_features.py

What you'll learn:

  • Mixed feature classification
  • Categorical feature configuration
  • Feature importance analysis
  • Model comparison techniques

4. Advanced Training

Explores advanced training configurations:

  • Custom PyTorch Lightning trainer parameters
  • Different hardware configurations (CPU/GPU)
  • Training optimization techniques
  • Model comparison and selection

Run the example:

cd examples
uv run --extra huggingface python advanced_training.py

What you'll learn:

  • Advanced training configurations
  • Hardware-specific optimizations
  • Training parameter tuning
  • Model performance comparison

5. Simple Explainability

Demonstrates model explainability with ASCII histogram visualizations:

  • Training a FastText classifier with enhanced data
  • Word-level contribution analysis
  • ASCII histogram visualization in terminal
  • Interactive mode for custom text analysis
  • Real-time prediction explanations

Run the example:

cd examples
# Regular mode - analyze predefined examples
uv run --extra huggingface python simple_explainability_example.py

# Interactive mode - analyze your own text
uv run --extra huggingface python simple_explainability_example.py --interactive

What you'll learn:

  • Model explainability and interpretation
  • Word importance analysis
  • Interactive prediction tools
  • ASCII-based data visualization
  • Real-time model analysis

🚀 Quick Start

To run any example:

  1. Install dependencies:
uv sync
  1. Navigate to examples directory:
cd examples
  1. Run an example:
uv --extra huggingface run python basic_classification.py

📊 Example Outputs

Basic Classification

🚀 Basic Text Classification Example
==================================================
📝 Creating sample data...
Training samples: 10
Validation samples: 2
Test samples: 3

🏗️ Creating FastText classifier...
🔨 Building model...
✅ Model built successfully!

🎯 Training model...
✅ Training completed!

🔮 Making predictions...
Predictions: [1 0 1]
True labels: [1 0 1]
Test accuracy: 1.000

📊 Detailed Results:
----------------------------------------
1. ✅ Predicted: Positive
   Text: This is an amazing product with great features!...

2. ✅ Predicted: Negative
   Text: Completely disappointed with this purchase...

3. ✅ Predicted: Positive
   Text: Excellent build quality and works as expected...

💾 Saving model configuration...
✅ Configuration saved to 'basic_classifier_config.json'

🎉 Example completed successfully!

Multi-class Classification

🎭 Multi-class Text Classification Example
==================================================
📝 Creating multi-class sentiment data...
Training samples: 15
Class distribution: Negative=5, Neutral=5, Positive=5

🏗️ Creating multi-class FastText classifier...
🔨 Building model...
✅ Model built successfully!

🎯 Training model...
✅ Training completed!

📊 Detailed Results:
------------------------------------------------------------
1. ✅ Predicted: Negative, True: Negative
   Text: This is absolutely horrible!

2. ✅ Predicted: Neutral, True: Neutral
   Text: It's an average product, nothing more.

3. ✅ Predicted: Positive, True: Positive
   Text: Fantastic! Love every aspect of it!

Final Accuracy: 3/3 = 1.000

Simple Explainability

🔍 Simple Explainability Example

🔍 Testing explainability on 5 examples:
============================================================

📝 Example 1:
Text: 'This product is amazing!'
Prediction: Positive

📊 Word Contribution Histogram:
--------------------------------------------------
        This | ██████████████████████████████ 0.3549
     product | █████████████                  0.1651
          is | ████████████████████████       0.2844
    amazing! | ████████████████               0.1956
--------------------------------------------------
✅ Analysis completed for example 1

📝 Example 2:
Text: 'Poor quality and terrible service'
Prediction: Negative
⚠️  Explainability failed: 
✅ Analysis completed for example 2

📝 Example 3:
Text: 'Great value for money'
Prediction: Positive

📊 Word Contribution Histogram:
--------------------------------------------------
       Great | ██████████████████████████████ 0.3287
       value | ████████████████████           0.2220
         for | ██████████████████████████     0.2929
       money | ██████████████                 0.1564
--------------------------------------------------
✅ Analysis completed for example 3

📝 Example 4:
Text: 'Completely disappointing and awful experience'
Prediction: Negative

📊 Word Contribution Histogram:
--------------------------------------------------
  Completely | ██████████                     0.1673
disappointing | ██████████████████████████████ 0.4676
         and | █████                          0.0910
       awful | ███████                        0.1225
  experience | █████████                      0.1516
--------------------------------------------------
✅ Analysis completed for example 4

📝 Example 5:
Text: 'Love this excellent design'
Prediction: Positive

📊 Word Contribution Histogram:
--------------------------------------------------
        Love | ██████████████████             0.2330
        this | ████████████████████           0.2525
   excellent | ██████████████████████████████ 0.3698
      design | ███████████                    0.1447
--------------------------------------------------
✅ Analysis completed for example 5

🎉 Explainability analysis completed for 5 examples!

💡 Tip: Use --interactive flag to enter interactive mode for custom text analysis!
   Example: uv run python examples/simple_explainability_example.py --interactive

Interactive Explainability Mode

============================================================
🎯 Interactive Explainability Mode
============================================================
Enter your own text to see predictions and explanations!
Type 'quit' or 'exit' to end the session.

💬 Enter text: Amazing product quality!

🔍 Analyzing: 'Amazing product quality!'
🎯 Prediction: Positive

📊 Word Contribution Histogram:
--------------------------------------------------
     Amazing | ██████████████████████████████ 0.5429
     product | ██████████████                 0.2685
    quality! | ██████████                     0.1886
--------------------------------------------------
💡 Most influential word: 'Amazing' (score: 0.5429)

--------------------------------------------------
💬 Enter text: Terrible customer support

🔍 Analyzing: 'Terrible customer support'
🎯 Prediction: Negative

📊 Word Contribution Histogram:
--------------------------------------------------
    Terrible | ██████████████████████████████ 0.5238
    customer | ███████████                    0.1988
     support | ███████████████                0.2774
--------------------------------------------------
💡 Most influential word: 'Terrible' (score: 0.5238)

--------------------------------------------------
💬 Enter text: quit
👋 Thanks for using the explainability tool!

🛠️ Customizing Examples

Modify Data

You can easily adapt the examples to your own data:

# Replace the example data with your own
X_train = np.array([
    "Your text sample 1",
    "Your text sample 2",
    # ... more samples
])
y_train = np.array([0, 1, ...])  # Your labels

Adjust Parameters

Experiment with different model parameters:

classifier = create_fasttext(
    embedding_dim=200,    # Increase for better representations
    num_tokens=20000,     # Increase for larger vocabularies
    min_count=3,          # Increase to filter rare words
    num_epochs=100,       # Increase for more training
    batch_size=64,        # Adjust based on your hardware
)

Add Custom Features

Extend examples with your own categorical features:

# Add your categorical features
categorical_features = np.array([
    [category1, category2, category3],
    # ... more feature vectors
])

X_mixed = np.column_stack([text_data, categorical_features])

📈 Performance Tips

For Better Accuracy

  1. Increase embedding dimensions for complex tasks
  2. Use more training data when available
  3. Tune n-gram parameters (min_n, max_n) for your domain
  4. Experiment with batch sizes and learning rates
  5. Consider mixed features if you have structured data

For Faster Training

  1. Use sparse embeddings for large vocabularies
  2. Increase batch size (if memory allows)
  3. Reduce embedding dimensions for faster convergence
  4. Use CPU training for small datasets
  5. Adjust num_workers for optimal data loading

For Large Datasets

  1. Use gradient accumulation with small batch sizes
  2. Enable mixed precision training (precision=16)
  3. Implement data streaming for very large datasets
  4. Use multiple GPUs if available

🐛 Troubleshooting

Common Issues

  1. Memory errors:

    • Reduce batch_size
    • Use sparse=True
    • Reduce embedding_dim

  2. Slow training:

    • Increase batch_size
    • Reduce num_workers
    • Use CPU for small datasets

  3. Poor accuracy:

    • Increase training data
    • Tune hyperparameters
    • Check data quality
    • Increase num_epochs

  4. Import errors:

    • Run uv sync to install dependencies
    • Check Python version compatibility

Getting Help

If you encounter issues:

  1. Check the main README for setup instructions
  2. Review the API documentation
  3. Look at similar examples for reference
  4. Open an issue on GitHub with your specific problem

🔗 Related Documentation

🤝 Contributing Examples

We welcome new examples! If you have a use case that would benefit others:

  1. Follow the existing example structure
  2. Include comprehensive comments
  3. Add error handling and validation
  4. Test your example thoroughly
  5. Update this README with your addition

Example template structure:

"""
Your Example Title

Brief description of what this example demonstrates.
"""

import numpy as np
from torchTextClassifiers import create_fasttext

def main():
    print("🚀 Your Example Title")
    print("=" * 50)
    
    # Your implementation here
    
    print("🎉 Example completed successfully!")

if __name__ == "__main__":
    main()