costlist = []
showevery = 10
for episode in range(numepisodes):
if episode % 100 == 0:
showevery = max(1, showevery - 1)
if episode % showevery == 0:
render = True
eps = None
print('episode', episode, 'l rate', agent.getlearnrate())
oldavg = avg
else:
render = False
eps = episode
startt = time.time()
total_rew, steps, cost, listob, listact = agents.do_rollout(
agent, env, eps, render=render)
if episode == 0:
avg = total_rew
if episode % 50 == 0:
print(agent.config)
if episode % 1 == 0:
listob = np.array(listob)
listact = np.array(listact)
allactionsparse = np.zeros((listact.shape[0], agent.n_out))
allactionsparse[np.arange(listact.shape[0]), listact] = 1.
inc = max(0.06, 1. / (episode + 1.)**0.6)
avg = avg * (1 - inc) + inc * total_rew
totrewlist.append(total_rew / agent.config['scalereward'])
totrewavglist.append(avg / agent.config['scalereward'])
totrewlist = []
totrewavglist = []
costlist = []
showevery = 10
for episode in range(numepisodes):
if episode % showevery == 0:
render = True
eps = None
print ('episode', episode, 'l rate', agent.getlearnrate())
oldavg = avg
else:
render = False
eps = episode
startt = time.time()
total_rew, steps, cost, listob, listact = agents.do_rollout(agent, env, eps, render=render)
if episode == 0:
avg = total_rew
if episode % 50 == 0:
print(agent.config)
if episode % 1 == 0:
listob = np.array(listob)
listact = np.array(listact)
allactionsparse = np.zeros((listact.shape[0], agent.n_out))
allactionsparse[np.arange(listact.shape[0]), listact] = 1.
inc = max(0.06, 1. / (episode + 1.) ** 0.6)
avg = avg * (1 - inc) + inc * total_rew
totrewlist.append(total_rew / agent.config['scalereward'])
totrewavglist.append(avg / agent.config['scalereward'])