This project predicts the number of available bikes in Montpellier, France, using a small machine learning model that can be executed directly in a web browser.
- Bike station data: Montpellier Open Data Portal
- Weather data: NOAA Global Summary of the Day
The model uses the following features for prediction:
['is_weekend', 'is_holiday', 'is_school_vacation', 'day_of_week', 'temp', 'max_temp', 'min_temp', 'precipitation', 'wind_speed', 'visibility', 'fog', 'rain', 'snow', 'hail', 'thunder', 'tornado', 'hour', 'minute']
Note: Some features (e.g., tornado) may not be particularly relevant for Montpellier but are included for model completeness.
- Framework: PyTorch
- Web Implementation: ONNX Runtime for browser-based execution
The bike_predictor_web directory contains the implementation of the web-based prediction interface.
Based on 8000 test samples:
- Mean Absolute Error: 0.38
- Maximum Error: 3.58
- Actual Value: 8.00
- Predicted Value: 4.42
- Number of errors > 1: 428
- Number of errors ≤ 1: 7572
- Percentage of errors > 1: 5.35%
Sample predictions:
- Actual: 5.00, Predicted: 4.83, Error: 0.17
- Actual: 6.00, Predicted: 5.75, Error: 0.25
- Actual: 6.00, Predicted: 5.68, Error: 0.32
- Actual: 4.00, Predicted: 3.94, Error: 0.06
- Actual: 6.00, Predicted: 5.64, Error: 0.36
To test the model:
-
Install Flask:
pip install flask -
Run the web application:
python bike_predictor_web/app.py -
Open a web browser and navigate to
http://localhost:5000 -
Use the following sample data to test the model:
- Day: Tuesday (day_of_week: 2)
- Is weekend: No
- Is holiday: No
- Is school vacation: Yes
- Temperature: 22.5°C
- Max Temperature: 27.4°C
- Min Temperature: 18.5°C
- Precipitation: 0.0 mm
- Wind Speed: 7.9 m/s
- Visibility: 11.9 km
- Fog, Rain, Snow, Hail, Thunder, Tornado: No
- Time: 10:15
With these inputs, the model should return a prediction close to 8 available bikes.