Seq2Key: A Predictive Keyboard Model

A deep learning-based predictive keyboard application with LSTM neural networks

📋 Table of Contents

• ✨ Features
• 🛠️ Requirements
• 🚀 Setup
• 💻 Usage
• 📚 Model Architecture
• 📁 File Structure
• 📄 License

✨ Features

• 🧠 LSTM-based predictive text model for accurate next-word predictions
• ⚡ Real-time word prediction with instant results
• 🎯 Top-k sampling for diverse and relevant predictions
• 🌙 Dark mode UI for comfortable viewing in any lighting
• 🖥️ Interactive web interface built with Streamlit
• 📊 Confidence scores for each prediction
• 📈 Configurable prediction count to control output

🛠️ Requirements

Python Dependencies

• Python 3.7+
• PyTorch >= 1.9.0
• Streamlit >= 1.28.0
• NumPy >= 1.21.0

System Requirements

• Modern CPU (GPU recommended for training)
• 2GB+ RAM
• 50MB+ disk space

🚀 Setup

1. Clone the Repository

git clone https://github.com/BLShaw/Seq2Key
cd Seq2Key

2. Install Dependencies

pip install -r requirements.txt

3. Prepare Model File

You need the model file predictive_keyboard_model.pth in the project directory. This file is typically too large for Git repositories. If it's not included, you'll need to train your own model using the provided notebook or download a pre-trained model file.

To train your own model, run the predictive_keyboard_model.ipynb notebook with your training data (TXT).

4. Run the Application

streamlit run app.py

💻 Usage

1. Input Text: Enter your text in the input field (e.g., "the quick brown")
2. Configure Predictions: Adjust the number of predictions to show using the slider
3. Get Predictions: Click "Get Predictions" to see the forecasted words
4. View Results: The model displays the top-k most likely next words with confidence scores

Example Usage

• Input: "the quick"
• Predictions: ["brown", "way", "and", "time", "man"]
• Confidence Scores: [0.45, 0.23, 0.15, 0.12, 0.05]

📚 Model Architecture

The predictive model utilizes a sophisticated LSTM-based architecture:

• Embedding Layer: Converts tokens to dense vector representations
• LSTM Layers: Two-layer LSTM for sequence processing and context understanding
• Dropout Layer: Prevents overfitting with 30% dropout rate
• Fully Connected Layer: Outputs probability distribution over vocabulary
• Special Tokens: Support for padding, unknown, start-of-sequence, and end-of-sequence tokens

Neural Network Specifications

Component	Details
Embedding Dimension	128
Hidden Dimension	256
Number of Layers	2
Dropout Rate	0.3
Vocabulary Size	4500+

📁 File Structure

predictive-keyboard-model/
├── app.py                 # Main Streamlit application
├── predictive_keyboard_model.pth   # Pre-trained model weights
├── predictive_keyboard_model.ipynb # Training notebook
├── dataset.txt            # Training data
├── requirements.txt       # Python dependencies
├── README.md             # Project documentation
├── .gitignore           # Git ignore rules
└── LICENSE              # License information

Key Files Description

• app.py: Interactive web application with dark mode UI
• predictive_keyboard_model.pth: Trained model weights and vocabulary
• predictive_keyboard_model.ipynb: Training notebook with detailed implementation
• dataset.txt: Text corpus for training the model
• requirements.txt: List of required Python packages

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

Made with ❤️ using PyTorch and Streamlit

⭐ Star this repository if you found it helpful!