search

gijskoning / Reproducibility_project

3 stars 0 forks source link

research first model #1

Open gijskoning opened 3 years ago

gijskoning commented 3 years ago

Might be good to first start with only the FNN. Also found out that it is better to start working with the Traffic Control environment since that model is a lot smaller.

gijskoning commented 3 years ago

image image

Table 4. PPO hyperparameters.

Parameter value
learning rate 2.5e-4
discount y 0.99
GAE delta 0.95
memory size 128
batch size 32
num. epoch 3
num. workers 8
entropy B 1.0e-2
clip e 0.1
value coeff. c1 1