Handwritten Digit Recognition (KNN) — Streamlit App

Small Streamlit app that lets you draw a digit (0–9) and predicts it using a KNN model.

Project layout

• app.py — Streamlit frontend
• knn_digit_model.joblib — trained KNN model (expected in project root)
• scaler.joblib — feature scaler used before prediction (expected in project root)
• requirements.txt — Python dependencies

Requirements

Install dependencies (Windows):

python -m venv .venv
.\.venv\Scripts\Activate
pip install -r requirements.txt

If opencv-python fails to install, ensure you are using a supported Python version and run:

pip install opencv-python

Run the app

From project root:

.\.venv\Scripts\Activate
streamlit run app.py

Open the displayed local URL in your browser.

How to use

1. Draw a digit (0–9) on the canvas.
2. The app preprocesses the canvas image to an 8x8 feature vector, scales it with scaler.joblib, and predicts with knn_digit_model.joblib.
3. The predicted digit is shown below the canvas.

Notes & troubleshooting

• Make sure knn_digit_model.joblib and scaler.joblib are in the same folder as app.py. If you used different filenames or a subfolder, update the paths in app.py.
• The canvas returns an RGBA image; the app converts it to grayscale before resizing. If predictions seem wrong, you may need to adjust preprocessing (resize method, inversion/scaling) to match how the model was trained.
• If you get errors loading the joblib files, confirm they were saved with the same scikit-learn / joblib versions used here.
• Use opencv-python (not cv2-python) in requirements if installation issues occur.

Re-training a model (brief)

If you want to retrain using scikit-learn's digits dataset:

• Load dataset from sklearn.datasets.load_digits()
• Preprocess (flatten 8x8 images), fit a StandardScaler, then a KNeighborsClassifier
• Save with joblib:

import joblib
joblib.dump(model, "knn_digit_model.joblib")
joblib.dump(scaler, "scaler.joblib")

License

MIT