else:
vis.set_gp_cartpole_state(rollout_gp[i - 1][3],
rollout_gp[i - 1][2])
vis.set_gp_cartpole_rollout_state(
rollout_gp_trajs[:, i - 1, 3], rollout_gp_trajs[:, i - 1,
2])
vis.set_gp_delta_state_trajectory(ts[:i + 1], pred_gp_mean[:i + 1],
pred_gp_variance[:i + 1])
if policy == swingup_policy:
policy_type = 'swing up'
else:
policy_type = 'random'
vis.set_info_text('epoch: %d\npolicy: %s' % (epoch, policy_type))
vis_img = vis.draw(redraw=(i == 0))
cv2.imshow('vis', vis_img)
if epoch == 0 and i == 0:
# First frame
video_out = cv2.VideoWriter(
'cartpole.mp4', cv2.VideoWriter_fourcc('m', 'p', '4', 'v'),
int(1.0 / DELTA_T), (vis_img.shape[1], vis_img.shape[0]))
video_out.write(vis_img)
cv2.waitKey(int(1000 * DELTA_T))
# Augment training data
new_train_x, new_train_y = make_training_data(state_traj[:-1],
else:
vis.set_gp_cartpole_state(rollout_gp[i - 1][3],
rollout_gp[i - 1][2])
vis.set_gp_cartpole_rollout_state(
rollout_gp_trajs[:, i - 1, 3], rollout_gp_trajs[:, i - 1,
2])
vis.set_gp_delta_state_trajectory(ts[:i + 1], pred_gp_mean[:i + 1],
pred_gp_variance[:i + 1])
if policy == swingup_policy:
policy_type = "swing up"
else:
policy_type = "random"
vis.set_info_text("epoch: %d\npolicy: %s" % (epoch, policy_type))
vis_img = vis.draw(redraw=(i == 0))
cv2.imshow("vis", vis_img)
if epoch == 0 and i == 0:
# First frame
video_out = cv2.VideoWriter(
"cartpole.mp4",
cv2.VideoWriter_fourcc("m", "p", "4", "v"),
int(1.0 / DELTA_T),
(vis_img.shape[1], vis_img.shape[0]),
)
video_out.write(vis_img)
cv2.waitKey(int(1000 * DELTA_T))