PriceLens

Estimate the market value of an Edmonton home from its size, layout and age.

https://pricelens.streamlit.app

Features

• Price estimates from four inputs: size, bedrooms, bathrooms, year built
• A user-friendly Streamlit interface for quick estimates.
• Input bounds match the training data range
• Model parameters saved after training, so predictions are instant

Model

Multivariable linear regression trained with gradient descent, implemented from scratch in NumPy. Features: square footage, bedrooms, bathrooms and year built. Inputs are z-score normalized using the training set's mean and standard deviation.

Trained on 1,653 Edmonton listings after outlier filtering.

Tech stack

• Frontend: Streamlit
• Backend: Python
• Data Processing: pandas, NumPy
• Training Diagnostics: Matplotlib

Run locally

git clone https://github.com/lxeu/PriceLens.git
cd PriceLens
pip install -r requirements.txt
streamlit run app.py

To retrain the model from the raw data:

python src/train.py

Related

The model is developed in edmonton-housing-price-predictor. I started from a single-variable regression and built up to the multivariable model used here.

Limitations

• No sense of location, which is the single most impactful feature that is missing.
• Systematically underestimates homes above roughly $1M
• Trained on a single market snapshot

Roadmap

• Neighbourhood as a feature
• Vectorized gradient descent
• Half baths and garages as features