def test_policy2():
from train_script import built_mixedpg_parser
import gym
args = built_mixedpg_parser()
env = gym.make('Pendulum-v0')
policy_with_value = PolicyWithQs(env.observation_space, env.action_space,
args)
def test_trained_model(model_dir, ppc_params_dir, iteration):
from train_script import built_mixedpg_parser
from policy import PolicyWithQs
args = built_mixedpg_parser()
evaluator = Evaluator(PolicyWithQs, args.env_id, args)
evaluator.load_weights(model_dir, iteration)
evaluator.load_ppc_params(ppc_params_dir)
return evaluator.metrics(1000, render=False, reset=False)
def test_policy():
import gym
from train_script import built_mixedpg_parser
args = built_mixedpg_parser()
print(args.obs_dim, args.act_dim)
env = gym.make('PathTracking-v0')
policy = PolicyWithQs(env.observation_space, env.action_space, args)
obs = np.random.random((128, 6))
act = np.random.random((128, 2))
Qs = policy.compute_Qs(obs, act)
print(Qs)
def test_policy_with_Qs():
from train_script import built_mixedpg_parser
import gym
import numpy as np
import tensorflow as tf
args = built_mixedpg_parser()
args.obs_dim = 3
env = gym.make('Pendulum-v0')
policy_with_value = PolicyWithQs(env.observation_space, env.action_space, args)
# print(policy_with_value.policy.trainable_weights)
# print(policy_with_value.Qs[0].trainable_weights)
obses = np.array([[1., 2., 3.], [3., 4., 5.]], dtype=np.float32)
with tf.GradientTape() as tape:
acts, _ = policy_with_value.compute_action(obses)
Qs = policy_with_value.compute_Qs(obses, acts)[0]
print(Qs)
loss = tf.reduce_mean(Qs)
gradient = tape.gradient(loss, policy_with_value.policy.trainable_weights)
print(gradient)