search

cs534-propagators / kaggle-lyft-motion-prediction

WPI CS534 Term Project - Lyft Motion Prediction for Autonomous Vehicles where we build motion prediction models for self-driving vehicles
https://www.kaggle.com/c/lyft-motion-prediction-autonomous-vehicles
4 stars 0 forks source link

Compare sliding vs non-sliding window training sets creation #13

Open yahelnachum opened 4 years ago

yahelnachum commented 4 years ago
Resolution Table Date Branch Link Details
- - - -
- - - -

Problem

We're not sure what effects reusing training data has on the model's ability to predict.

Solution

Compare a sliding vs non-sliding window training set generation and its effect on a model's ability to predict future data.

Additional Info

Right now we are using a sliding window to create a lot of training data.

Example:

  1. a single training agent might have 15 frames of history
  2. using a window of size 10 will give us 5 training sets (10 history (input), 1 future (output)) 
    1-10 input, 11 output
2-11 input, 12 output
3-12 input, 13 output
4-13 input, 14 output
5-14 input, 15 output