from ddpg import Agent
import gym
import numpy as np
env = gym.make('LunarLanderContinuous-v2')
agent = Agent(alpha = 0.000025, beta = 0.00025, input_dims = [8], tau = 0.001, env = env, batch_size = 64, layer1_size = 400, layer2_size = 300, n_actions = 2)
np.random.seed(42)
score_history = []
for i in range(1000):
done = False
score = 0
obs = env.reset()
while not done:
act = agent.choose_action(obs)
new_state, reward, done, info = env.step(act)
agent.remember(obs, act, reward, new_state, int(done))
agent.learn()
score += reward
obs = new_state
score_history.append(score)
print("Episode - {} Score - {} 100 game average {}".format(i, score, np.mean(score_history[-100:])))
if i % 25 == 0:
agent.save_models()
filename = l
normalized.append(reward)
if reward > best_parameters[1]:
best_parameters[0] = action
best_parameters[1] = reward
best_parameters[2] = step*(episode + 1)
all_steps[episode] = (step*episode+step)
exit = True
rewards.append(reward)
if reward > -10000:
inlook.append(reward)
avg_rewards.append(np.mean(inlook[-10:]))
if reward > 0:
done = True
env.render()
agent.learn(batch_size)
state = new_state
inlook2 = []
avg_rewards2 = []
for episode in range(1):
noise.reset()
eps_reward = 0
print(episode)
best_parameters = [(0,0,0), 0, 0]
step = 0
for i in range(total_steps):
print(step)
step += 1
setpoint = 20 if random == False else np.random.random()*100