Distribution difference between Observations and Predicted Actions

sisl / ngsim_env

Learning human driver models from NGSIM data with imitation learning.

https://arxiv.org/abs/1803.01044

MIT License

172 stars 80 forks source link

Distribution difference between Observations and Predicted Actions #6

Closed HighlyAuditory closed 5 years ago

HighlyAuditory commented 5 years ago

The 'turn_rate_global' feature in the 66-dim observation has max, min, mean approximately 0.05, -0.05, 0. While I generated some trajectories with PS-GAIL and get the 'mean' action output with statistics max, min, mean as 1.18, -1.36 -0.07. Why are the two distributions different? Are they with the same units?

djp42 commented 5 years ago

Can you share more details about how you are calculating these statistics, how many trajectories you are generating, etc? The units should generally be the same, and I can provide more insight if I have more details about what you are doing.

HighlyAuditory commented 5 years ago

You were right. The number of my sampled trajs is too small. Thank you!

djp42 commented 5 years ago

Great, glad we resolved it!