Neural network to predict taxi destination
This work aims at resolving a Kaggle Competition to predict the destination of taxi trips based on initial partial trajectories: https://www.kaggle.com/c/pkdd-15-predict-taxi-service-trajectory-i/overview.
This project is composed of four notebook files:
- Data_visualisation.ipynb : it visualizes data to prepare the feature engineering
- Tensoring.ipynb : it pre-processes data to put them in tensor form, and adapt them to our network.
- Network_taxi.ipynb : it defines our neural network and traines it.
- Test_model.ipynb : it computes results for the test dataset.
The project is also composed of :
- a file "model_cluster_100.pth" that stores the weights of our network after the 100th epoch
- a file "loss_cluster.txt" that stores the loss obtained after each epoch while training our network
To run our project, you have to :
- download data from the kaggle competition website: https://www.kaggle.com/c/pkdd-15-predict-taxi-service-trajectory-i/data
- run all the Data_viz.ipynb notebook (data are then ready to be pre-processed)
- run all the Tensoring.ipynb notebook (data are then ready to be used in our neural network)
- run all the Network_taxi.ipynb notebook (weights of the neural networks are saved into pth files each two epochs and losses are saved into a txt file)
- run all the Test_model.ipynb notebook
This project has been carried out by Alodie Boissonnet, Jean Bouteiller and Barnabé Mas, under the supervision of Marc Lelarge.