FrozenLake with PPO

In this notebook we solve a non-slippery version of the FrozenLake-v0 environment using the TD actor critic algorithm with PPO policy updates.

We’ll use a linear function approximator for our policy \(\pi_\theta(a|s)\) and our state value function \(v_\theta(s)\). Since the observation space is discrete, this is equivalent to the table-lookup case.

---

ppo.py

import coax
import jax
import jax.numpy as jnp
import gymnasium
import haiku as hk
import optax


# the MDP
env = gymnasium.make('FrozenLakeNonSlippery-v0')
env = coax.wrappers.TrainMonitor(env)


def func_v(S, is_training):
    value = hk.Sequential((hk.Linear(1, w_init=jnp.zeros), jnp.ravel))
    return value(S)


def func_pi(S, is_training):
    logits = hk.Linear(env.action_space.n, w_init=jnp.zeros)
    return {'logits': logits(S)}


# function approximators
pi = coax.Policy(func_pi, env)
v = coax.V(func_v, env)


# create copies
pi_old = pi.copy()  # behavior policy
v_targ = v.copy()   # target network


# experience tracer
tracer = coax.reward_tracing.NStep(n=1, gamma=0.9)


# updaters
simple_td = coax.td_learning.SimpleTD(v, v_targ, optimizer=optax.adam(0.02))
ppo_clip = coax.policy_objectives.PPOClip(pi, optimizer=optax.adam(0.01))


# train
for ep in range(500):
    s, info = env.reset()

    for t in range(env.spec.max_episode_steps):
        a, logp = pi_old(s, return_logp=True)
        s_next, r, done, truncated, info = env.step(a)

        # small incentive to keep moving
        if jnp.array_equal(s_next, s):
            r = -0.01

        # update
        tracer.add(s, a, r, done, logp)
        while tracer:
            transition_batch = tracer.pop()
            _, td_error = simple_td.update(transition_batch, return_td_error=True)
            ppo_clip.update(transition_batch, td_error)

            # sync target networks
            v_targ.soft_update(v, tau=0.01)
            pi_old.soft_update(pi, tau=0.01)

        if done or truncated:
            break

        s = s_next

    # early stopping
    if env.avg_G > env.spec.reward_threshold:
        break


# run env one more time to render
s, info = env.reset()
env.render()

for t in range(env.spec.max_episode_steps):

    # estimated state value
    print("  v(s) = {:.3f}".format(v(s)))

    # print individual action probabilities
    params = pi.dist_params(s)
    propensities = jax.nn.softmax(params['logits'])
    for i, p in enumerate(propensities):
        print("  π({:s}|s) = {:.3f}".format('LDRU'[i], p))

    a = pi.mode(s)
    s, r, done, truncated, info = env.step(a)

    env.render()

    if done or truncated:
        break


if env.avg_G < env.spec.reward_threshold:
    name = globals().get('__file__', 'this script')
    raise RuntimeError(f"{name} failed to reach env.spec.reward_threshold")