Part of our code base for combining a/rqmc with pg methods.

Files of interest:

• main.py: main entry point for learning (not only LQR, but for all environments and methods).
• rqmc_distributions/normal_rqmc.py and rqmc_distributions/uniform_rqmc.py: define PyTorch Distributions classes for RQMC sampling, built on top of SSJ (also see rqmc_distributions/ssj_sobol.py which defines the SSJ Java → Python wrapper).
• envs/lqr.py: implementation of the LQR environment, which is typically instantiated as follows:

lqr=LQR(
N=6,
M=8,
init_scale=3.0,
max_steps=20,
Sigma_s_kappa=1.0,
Q_kappa=3.0,
P_kappa=3.0,
A_norm=1.0,
B_norm=1.0,
Sigma_s_scale=1.0,
random_init=False,
lims=100,
)