Any getting started notebook?

[RajS999]

Is there any notebook / code snippet giving minimal example, something like this for stable-baselines3:

import gym

from stable_baselines3 import A2C

env = gym.make('CartPole-v1')

model = A2C('MlpPolicy', env, verbose=1)
model.learn(total_timesteps=10000)

obs = env.reset()
for i in range(1000):
    action, _state = model.predict(obs, deterministic=True)
    obs, reward, done, info = env.step(action)
    env.render()
    if done:
      obs = env.reset()

Is such usage of interact library possible?

Or I should go through interact.train.py to figure out how can I use this library in code, instead of calling it from command line as explained in readme page.

[rystrauss]

There are no example notebooks, but train.py and play.py both contain most of what you'd need to know. Every agent implements the same interface (see the Agent class in interact.agents.base), so it's really just a matter of calling those methods. I really only wrote the code with the intention of using the command-line interface though, so that's why there aren't any code examples.

[RajS999]

Thanks for the reply. Just one more question, have you benchmarked / evaluated these algorithms? May be for performance or correctness? May be am asking for too much... 😶

[rystrauss]

I have not formally and thoroughly benchmarked them. I tested them all on one or two environments to make sure they were in fact learning, but I can't guarantee there aren't any subtle bugs that would would keep these implementations from matching published results. These implementations are mostly for my own edification and are not intended to be a well-vetted "product" like Stable Baselines or RLlib.