Predictive Text Generator

A lightweight N-gram based predictive text generator with multiple zero-dependency interfaces:

• Local Web UI using only Python's standard library (no external packages)
• Desktop UI using Tkinter (standard library)
• Command Line Interface (CLI)
• Optional Streamlit UI (requires dependencies)

Features

• Robust N-gram model (configurable N from 2–4)
• Preprocessing and tokenization with NLTK when available, with a built-in regex fallback
• Laplace smoothing (configurable k)
• Interpolated prediction combining different N-gram orders
• User dictionary boosts: global word boosts and context-specific boosts
• Model persistence: save to / load from JSON

Project Structure

.
├── app.py            # Streamlit UI (optional)
├── cli.py            # Command Line Interface
├── ngram_model.py    # Core N-gram model and utilities
├── tk_app.py         # Tkinter Desktop UI (fallback visual interface)
├── web_server.py     # Zero-dependency local Web UI (standard library)
├── user_dict.json    # Optional user boosts file
├── requirements.txt  # Dependencies for Streamlit UI
└── model.json        # Example persisted model file

Quickstart (No Dependencies)

You can use the Web UI, Desktop UI, or CLI without any external packages.

1) Web UI (recommended)

• Start the server:
  • python web_server.py
• Open in your browser:
  • http://127.0.0.1:8000/
• Use the controls:
  • Parameters: set N-gram, Laplace k, Top-K, then click "Apply"
  • Corpus: paste text and click "Ingest Corpus"
  • Compose: type in the input field to see suggestions; click a suggestion pill to append
  • Boosts:
    • Global: word + value, click "Add"
    • Context: context + word + value, click "Add"
  • Model Persistence:
    • Path: set file path (e.g., model.json)
    • "Save" / "Load"

2) Desktop UI (Tkinter)

• Launch: python tk_app.py
• Similar controls for parameters, corpus ingestion, compose area, boosts, and save/load
• Note: Tkinter is part of the standard library, but on some systems may require installing "tk" support

3) CLI

• Run: python cli.py --n 3 --top-k 5
• Options:
  • --n N-gram order (2–4)
  • --laplace-k Laplace smoothing constant (default 1.0)
  • --top-k number of suggestions to show (default 5)
  • --corpus PATH (optional) load text file(s) into the model
• Interactive usage:
  • Type text and press Enter to see suggestions
  • Built-in help is available within the CLI (type help) and includes commands for global/context boosts and saving/loading the model

Optional: Streamlit UI

If you prefer Streamlit and can install dependencies:

1. Create/Use a Python virtual environment (already present in the project as .venv).
2. Upgrade pip (example for this project where the venv uses bin):
   • .venv\bin\python.exe -m pip install --upgrade pip
3. Install dependencies:
   • pip install -r requirements.txt
4. Run Streamlit:
   • python -m streamlit run app.py

If you encounter SSL/certificate or proxy issues when installing:

• Configure pip to use your corporate proxy and certificate store
• Try installing only streamlit first: pip install streamlit
• Prefer precompiled wheels for packages like numpy and pandas

Model Persistence

• Save: choose a path (e.g., model.json) in the UI or CLI command
• Load: choose the same path to restore
• The model JSON contains N, smoothing, counts, and user boosts

Tokenization Notes

• NLTK-based tokenization is used when available
• If NLTK is unavailable or punkt data is missing, the model automatically falls back to a regex tokenizer

Troubleshooting

• SyntaxError: bytes can only contain ASCII literal characters
  • Ensure HTML strings use Unicode (no b""") in web_server.py
• Tkinter not found
  • Install Python with Tk support or your OS-specific Tk package
• Streamlit / pip install fails (SSL, proxies)
  • Use your system/corporate proxy settings, trusted certificates, or precompiled wheels

License

• Internal/learning project. Add your preferred license as needed.