Ensembling function for multi-trajectory motion prediction for autonomous vehicles

This small repository is a follow-up to Lyft Kaggle competition on motion prediction:

link to competition

The code is described in some detail here: Kaggle forum post

There is also a discussion of the results here.

The Ensembling.ipynb notebook contains two functions

• ensembleTorch() iterative procedure for ensembling which converges to at least a local minimum
• run_ensemble() takes several individual model submission files and feeds them to the first function

Running of this notebook requires individual model submission files. The first and 9th place shared them in these kaggle datasets:

• 1st place
• 9th place