There are a lot of packages out there for deep Reinforcement Learning (RL), and so it became obvious to me that there needed to be yet another implementation! I’m calling it yarlp, Yet Another Reinforcement Learning Package! I’m developing it mostly for educational purposes.

Here are several great implementations that I’ve learned and borrowed from:

  1. rllab
  2. baselines
  3. mindpark

yarlp heavily uses tensorflow==1.3 and gym, and is meant to be simple to run, yet very configurable. Here is an example running TRPO:

from yarlp.agent.trpo_agent import TRPOAgent
from yarlp.utils.env_utils import NormalizedGymEnv
from yarlp.model.networks import mlp

env = NormalizedGymEnv(

agent = TRPOAgent(
    env, discount_factor=0.99,

agent.train(100, 0, n_steps=1024)

I’m planning to benchmark TRPO on several environments and then move over to value-based methods.

Here are tons of other RL packages:

comments powered by Disqus
Blog Logo

Baruch Tabanpour