This is a command line tool written in Python to train models for predicting house prices and a dockerized Flask web app to deploy the model.
The dependencies are listed in requirements.txt, and can be automatically installed using:
pip install -r requirements.txt.
Ideally do this in a fresh virtual environment.
After downloading the repo you may either run the flask web app to try out the model predictions or retrain the model using the command line interface.
To build the flask app you must have docker installed. Once you have installed docker, from the root directory run docker compose up and navigate in a web browser to the address mentioned, typically this is 127.0. 0.1:5000. You should be greeted by a web form which you can use to request predictions from the model.
To use the CLI to retrain the model, open a terminal in the root housepricer directory. Then you may run the following commands:
Run: python3 housepricer --data_dir data/ --model_dir data/ --hyperparameter_search random --iterations 150 for an example of setting hyperparameters with a random search
Run: python3 housepricer --data_dir data/ --model_dir data/ --hyperparameter_search evolve --population 50 --generations 25 for an example of setting hyperparameters with an evolutionary algorithm
If you want to run the californian test dataset use the flag --load_cal True for example you could run:
python3 housepricer --model_dir data/ --load_cal True --hyperparameter_search evolve --population 100 --generations 50
If you want to use the sklearn Histogram-based Gradient Boosting Regression Tree estimator instead of Random forest run
python3 housepricer --data_dir data/ --model_dir data/ --model_type notrandom --load_cal False --hyperparameter_search evolve --population 100 --generations 50
Note currently the included Bristol (UK) dataset is only from the freely available LandRegistry, which does not provide key features such as the number of rooms, size, or overall state of a property. The sklearn Californian dataset include both number of rooms and area of each property, so can be trained to a higher r2 value.