def finish(self, nb_train_steps, nb_epoch_steps, nb_warmup_steps=0):
import matplotlib
matplotlib.use('Agg')
from matplotlib import pyplot as plt
from util import mstd
x_vals = range(0, nb_train_steps + 1, nb_epoch_steps)
# Plot online reward and test reward.
for (k, v) in self.records.items():
v = np.array(v)
if len(v.shape) == 3:
for i in range(v.shape[1]):
f0, ax0 = plt.subplots()
m, ids25, ids75 = mstd(v[:, i, :].T)
# If the number of warmup steps > the number of epoch
# step or there is no episode completed, v may not have
# the same length with the x_vals.
_ = ax0.plot(x_vals[-len(v):], m, color='b')
_ = ax0.fill_between(x_vals[-len(v):],
list(ids75),
list(ids25),
facecolor='b',
alpha=0.2)
if k == 'mean_nlogdetcov':
_ = ax0.set_ylim(-1500, 3000)
_ = ax0.grid()
_ = ax0.set_ylabel(k)
_ = ax0.set_xlabel('Learning Steps')
if x_vals[-len(v)] < nb_warmup_steps:
_ = ax0.axvline(x=nb_warmup_steps, color='k')
_ = f0.savefig(
os.path.join(self.save_dir, "%s_eval_%d.png" % (k, i)))
elif len(v) > 0 and (k == 'test_reward' or k == 'mean_nlogdetcov'
or k == 'online_reward'):
f0, ax0 = plt.subplots()
if len(v.shape) == 2:
m, ids25, ids75 = mstd(v.T)
# If the number of warmup steps > the number of epoch
# step or there is no episode completed, v may not have
# the same length with the x_vals.
_ = ax0.plot(x_vals[-len(v):], m, color='b')
_ = ax0.fill_between(x_vals[-len(v):],
list(ids75),
list(ids25),
facecolor='b',
alpha=0.2)
else:
_ = ax0.plot(x_vals[-len(v):], v, color='b')
if k == 'mean_nlogdetcov':
_ = ax0.set_ylim(-1500, 3000)
_ = ax0.grid()
_ = ax0.set_ylabel(k)
_ = ax0.set_xlabel('Learning Steps')
if x_vals[-len(v)] < nb_warmup_steps:
_ = ax0.axvline(x=nb_warmup_steps, color='k')
_ = f0.savefig(os.path.join(self.save_dir, "%s.png" % k))
def batch_plot(list_records, save_dir, nb_train_steps, nb_epoch_steps, is_target_tracking=False):
import matplotlib
matplotlib.use('Agg')
from matplotlib import pyplot as plt
from util import mstd
results = {'online_reward':[], 'test_reward':[]}
if is_target_tracking:
results['mean_nlogdetcov'] = []
for r in list_records:
for (k,v) in results.items():
r[k] = np.array(r[k])
v.append(r[k].T)
x_vals = range(0, nb_train_steps+1, nb_epoch_steps)
for (k,v) in results.items():
v = np.array(v)
if len(v.shape) == 4:
for i in range(v.shape[2]):
v_i = np.concatenate(v[:,:,i,:], axis=0)
f0, ax0 = plt.subplots()
m, ids25, ids75 = mstd(v_i)
_ = ax0.plot(x_vals[-v_i.shape[1]:], m, color='k')
_ = ax0.fill_between(x_vals[-v_i.shape[1]:], list(ids75), list(ids25),
facecolor='k', alpha=0.2)
_ = ax0.plot(x_vals[-v_i.shape[1]:], np.max(v_i, axis=0), color='b')
_ = ax0.plot(x_vals[-v_i.shape[1]:], np.min(v_i, axis=0), color='r')
_ = ax0.grid()
if k == 'mean_nlogdetcov':
ax0.set_ylim(-1500, 3000)
_ = f0.savefig(os.path.join(save_dir, "%s_eval_%d.png"%(k,i)))
plt.close()
else:
if len(v.shape) == 3:
v = np.concatenate(v, axis=0)
f0, ax0 = plt.subplots()
m, ids25, ids75 = mstd(v)
_ = ax0.plot(x_vals[-v.shape[1]:], m, color='k')
_ = ax0.fill_between(x_vals[-v.shape[1]:], list(ids75), list(ids25),
facecolor='k', alpha=0.2)
_ = ax0.plot(x_vals[-v.shape[1]:], np.max(v, axis=0), color='b')
_ = ax0.plot(x_vals[-v.shape[1]:], np.min(v, axis=0), color='r')
_ = ax0.grid()
if k == 'mean_nlogdetcov':
ax0.set_ylim(-1500, 3000)
_ = f0.savefig(os.path.join(save_dir, k+".png"))
plt.close()