Pendulum with DDPG¶
In this notebook we solve the Pendulum environment using DDPG. We’ll use a simple multi-layer percentron for our function approximator for the policy and q-function.
This notebook periodically generates GIFs, so that we can inspect how the training is progressing.
After a few hundred episodes, this is what you can expect:
import gymnasium
import jax
import coax
import haiku as hk
import jax.numpy as jnp
from numpy import prod
import optax
# the name of this script
name = 'ddpg'
# the Pendulum MDP
env = gymnasium.make('Pendulum-v1', render_mode='rgb_array')
env = coax.wrappers.TrainMonitor(env, name=name, tensorboard_dir=f"./data/tensorboard/{name}")
def func_pi(S, is_training):
seq = hk.Sequential((
hk.Linear(8), jax.nn.relu,
hk.Linear(8), jax.nn.relu,
hk.Linear(8), jax.nn.relu,
hk.Linear(prod(env.action_space.shape), w_init=jnp.zeros),
hk.Reshape(env.action_space.shape),
))
mu = seq(S)
return {'mu': mu, 'logvar': jnp.full_like(mu, -10)} # (almost) deterministic
def func_q(S, A, is_training):
seq = hk.Sequential((
hk.Linear(8), jax.nn.relu,
hk.Linear(8), jax.nn.relu,
hk.Linear(8), jax.nn.relu,
hk.Linear(1, w_init=jnp.zeros), jnp.ravel
))
X = jnp.concatenate((S, A), axis=-1)
return seq(X)
# main function approximators
pi = coax.Policy(func_pi, env)
q = coax.Q(func_q, env, action_preprocessor=pi.proba_dist.preprocess_variate)
# target network
q_targ = q.copy()
pi_targ = pi.copy()
# experience tracer
tracer = coax.reward_tracing.NStep(n=5, gamma=0.9)
buffer = coax.experience_replay.SimpleReplayBuffer(capacity=25000)
# updaters
qlearning = coax.td_learning.QLearning(
q, pi_targ, q_targ, loss_function=coax.value_losses.mse, optimizer=optax.adam(1e-3))
determ_pg = coax.policy_objectives.DeterministicPG(pi, q_targ, optimizer=optax.adam(1e-4))
# action noise
noise = coax.utils.OrnsteinUhlenbeckNoise(mu=0., sigma=0.2, theta=0.15)
# train
while env.T < 1000000:
s, info = env.reset()
noise.reset()
noise.sigma *= 0.99 # slowly decrease noise scale
for t in range(env.spec.max_episode_steps):
a = noise(pi(s))
s_next, r, done, truncated, info = env.step(a)
# trace rewards and add transition to replay buffer
tracer.add(s, a, r, done)
while tracer:
buffer.add(tracer.pop())
# learn
if len(buffer) >= 5000:
transition_batch = buffer.sample(batch_size=128)
metrics = {'OrnsteinUhlenbeckNoise/sigma': noise.sigma}
metrics.update(determ_pg.update(transition_batch))
metrics.update(qlearning.update(transition_batch))
env.record_metrics(metrics)
# sync target networks
q_targ.soft_update(q, tau=0.001)
pi_targ.soft_update(pi, tau=0.001)
if done or truncated:
break
s = s_next
# generate an animated GIF to see what's going on
if env.period(name='generate_gif', T_period=10000) and env.T > 5000:
T = env.T - env.T % 10000 # round to 10000s
coax.utils.generate_gif(
env=env, policy=pi, filepath=f"./data/gifs/{name}/T{T:08d}.gif")