sisl/lstm-acc-predict

Analysis of Recurrent Neural Networks for Probabilistic Modeling of Driver Behavior

List of Files

• RNNspeed-predict.lua: used for training LSTM networks
• FFspeed-predict.lua: used for training feedforward networks

Sample command: th RNNspeed-predict.lua -mixture_size 2 -nn_size 128 -learning_rate 4e-3 -epochs 10 -dropout 0.25

• analyze.lua: propagates simulated trajectories using trained LSTM networks
• FF_analyze.lua: propagates simulated trajectories using trained feedforward networks
• The util folder contains various scripts called by the main scripts
• The model folder contains files that define the structure of the LSTM networks
• The analysis folder contains files that were used to propagate and analyze simulated trajectories. These include:
  • IDM_fit.ipynb: used to learn the parameters for the IDM model
  • compare_traj.jl: used to find RWSE, KL Divergence, and fraction of negative state values
  • cross_val.lua and ffcv.lua: scripts that load trained models and propagate simulated trajectories.

Note: The data used for this project is not included in this repository. The reconstructed NGSIM data can be obtained by following the instructions here.