Skip to content

jphalip/ECML-PKDD-2015

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Background

This repository contains code written for the Kaggle competition "ECML/PKDD 15: Taxi Trajectory Prediction". This implementation, based on Tensorflow and Keras, is inspired from the approach followed by the competition's winners. More information about this code is provided on this blog post.

For more information about the original competition winner's solution, please refer to:

Code structure

The code is comprised of three main files inside the code folder:

  • data.py: Methods for loading, cleaning and pre-processing the original datasets.
  • training.py: Methods for defining the neural network model and for running the training process.
  • utils.py: Various mathematical and graphical utility functions.

Getting started

This implementation is based on Tensforflow version 0.11.0. The training process for the included neural network model can be quite time-comsuming so it's recommended to use a GPU. A simple GPU-enabled Docker container setup can be found at: https://github.com/tensorflow/tensorflow/tree/master/tensorflow/tools/docker

Some extra python libraries, specified in requirements.txt file, will also have to be installed.

Once your environment is set up, download the competition's CSV data files in the datasets folder.

To run the training process, run the following:

    from code.training import full_train
    full_train(n_epochs=100, batch_size=200, save_prefix='mymodel')

The above will run the full training process and save some files to disk inside the cache folder:

  • mymodel-history.pickle as
  • 100 files (one for each epoch) named mymodel-XXX.hdf5 (with XXX replaced with each epoch number).

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages