def generate_place_test_data(demo_path, target_path, batch_size=1, index=0):
# color_num = ['color_blue', 'color_green', 'color_orange', 'color_yellow']
# object_num = ['object_type_animal', 'object_type_car', 'object_type_dinosaur', 'object_type_tool']
# print('demo_path', demo_path)
# print('target_path', target_path)
if FLAGS.demo_type == 'robot':
obsa, statea, actiona = read_data.Read_Robot_Data2(demo_path, FLAGS.T, index)
elif FLAGS.demo_type == 'human':
obsa, statea, actiona = read_data.Read_Human_Data2(demo_path, FLAGS.T, index)
obsb, stateb, actionb = read_data.Read_Robot_Data2(target_path, FLAGS.T, index)
obsas = np.reshape(obsa, [batch_size, FLAGS.T, FLAGS.im_width * FLAGS.im_height * FLAGS.num_channels])
obsbs = np.reshape(obsb, [batch_size, FLAGS.T, FLAGS.im_width * FLAGS.im_height * FLAGS.num_channels])
actionas = np.reshape(actiona, [batch_size, FLAGS.T, FLAGS.output_data])
actionbs = np.reshape(actionb, [batch_size, FLAGS.T, FLAGS.output_data])
# print('actionas', actionas)
# print('actionbs', actionbs)
stateas = np.zeros([batch_size, FLAGS.T, FLAGS.output_data])
statebs = np.zeros([batch_size, FLAGS.T, FLAGS.output_data])
return obsas, obsbs, actionas, actionbs, stateas, statebs
def generate_data(if_train=True):
if if_train:
batch_size = FLAGS.meta_batch_size
else:
batch_size = FLAGS.meta_test_batch_size
color_list = (np.random.randint(0, 100, size=batch_size) + 1) % FLAGS.color_num
print('color_list', color_list)
object_list = (np.random.randint(0, 100, size=batch_size) + 1) % FLAGS.object_num
print('object_list', object_list)
if if_train:
task_list = (np.random.randint(0, 100, size=batch_size) + 1) % FLAGS.train_task_num
else:
task_list = np.random.randint(FLAGS.train_task_num, FLAGS.task_num, size=batch_size)
print('task_list', task_list)
demo_list = (np.random.randint(0, 100, size=batch_size) + 1) % FLAGS.demo_num
print('demo_list', demo_list)
target_list = (np.random.randint(0, 100, size=batch_size) + 1) % FLAGS.demo_num
print('target_list', target_list)
obsas = []
obsbs = []
stateas = []
statebs = []
actionas = []
actionbs = []
color_num = ['color_blue', 'color_green', 'color_orange', 'color_yellow']
# color_num = ['color_blue', 'color_green', 'color_orange']
object_num = ['object_type_animal', 'object_type_car', 'object_type_dinosaur', 'object_type_tool']
for element in range(0, batch_size):
if if_train:
demo_path = '%s/%s/%s/%s/task_%d/demo_%d' % (
FLAGS.data_path, color_num[color_list[element]], object_num[object_list[element]], FLAGS.demo_type,
task_list[element], demo_list[element])
target_path = '%s/%s/%s/%s/task_%d/demo_%d' % (
FLAGS.data_path, color_num[color_list[element]], object_num[object_list[element]], FLAGS.target_type,
task_list[element], target_list[element])
else:
demo_path = '%s/%s/%s/%s/task_%d/demo_%d' % (
FLAGS.data_path, color_num[-1], object_num[object_list[element]], FLAGS.demo_type,
task_list[element], demo_list[element])
target_path = '%s/%s/%s/%s/task_%d/demo_%d' % (
FLAGS.data_path, color_num[-1], object_num[object_list[element]], FLAGS.target_type,
task_list[element], target_list[element])
# print('demo_path', demo_path)
# print('target_path', target_path)
index = np.random.randint(0, 20)
if FLAGS.demo_type == 'robot':
obsa, statea, actiona = read_data.Read_Robot_Data2(demo_path, FLAGS.T, index)
elif FLAGS.demo_type == 'human':
obsa, statea, actiona = read_data.Read_Human_Data2(demo_path, FLAGS.T, index)
obsb, stateb, actionb = read_data.Read_Robot_Data2(target_path, FLAGS.T, index)
obsas.append(obsa)
obsbs.append(obsb)
stateas.append(statea)
statebs.append(stateb)
actionas.append(actiona)
actionbs.append(actionb)
obsas = np.reshape(obsas, [batch_size, FLAGS.T, FLAGS.im_width * FLAGS.im_height * FLAGS.num_channels])
obsbs = np.reshape(obsbs, [batch_size, FLAGS.T, FLAGS.im_width * FLAGS.im_height * FLAGS.num_channels])
actionas = np.reshape(actionas, [batch_size, FLAGS.T, FLAGS.output_data])
actionbs = np.reshape(actionbs, [batch_size, FLAGS.T, FLAGS.output_data])
stateas = np.zeros([batch_size, FLAGS.T, FLAGS.output_data])
statebs = np.zeros([batch_size, FLAGS.T, FLAGS.output_data])
return obsas, obsbs, actionas, actionbs, stateas, statebs
def generate_atmaml_data(if_train=True):
if if_train:
batch_size = FLAGS.meta_batch_size
else:
batch_size = FLAGS.meta_test_batch_size
color_list = (np.random.randint(0, 100, size=batch_size) + 1) % FLAGS.color_num
# print('color_list', color_list)
compare_color_list = (color_list + np.random.randint(1, FLAGS.color_num-1)) % FLAGS.color_num
# print('compare_color_list', compare_color_list)
object_list = (np.random.randint(0, 100, size=batch_size) + 1) % FLAGS.object_num
# print('object_list', object_list)
compare_object_list = (object_list + np.random.randint(1, FLAGS.object_num-1)) % FLAGS.object_num
# print('compare_object_list', compare_object_list)
if if_train:
task_list = (np.random.randint(0, 100, size=batch_size) + 1) % FLAGS.train_task_num
compare_task_list = (task_list + np.random.randint(1, FLAGS.train_task_num)) % FLAGS.train_task_num
# compare_task_list = (task_list + np.random.randint(1, FLAGS.task_num - 1)) % FLAGS.task_num
else:
task_list = np.random.randint(FLAGS.train_task_num, FLAGS.task_num, size=batch_size)
compare_task_list = np.random.randint(FLAGS.train_task_num, FLAGS.task_num, size=batch_size)
# print('task_list', task_list)
# print('compare_task_list', compare_task_list)
demo_list = (np.random.randint(0, 100, size=batch_size) + 1) % FLAGS.demo_num
# print('demo_list', demo_list)
target_list = (np.random.randint(0, 100, size=batch_size) + 1) % FLAGS.demo_num
# print('target_list', target_list)
compare_demo_list = (task_list + np.random.randint(1, FLAGS.demo_num-1)) % FLAGS.demo_num
# print('compare_demo_list', compare_demo_list)
obsas = []
obsbs = []
obscs = []
extra_obses = []
stateas = []
statebs = []
statecs = []
extra_states = []
actionas = []
actionbs = []
actioncs = []
extra_actions = []
labelabs = []
labelcs = []
color_num = ['color_blue', 'color_green', 'color_orange', 'color_yellow']
# color_num = ['color_blue', 'color_green', 'color_orange']
object_num = ['object_type_animal', 'object_type_car', 'object_type_dinosaur', 'object_type_tool']
for element in range(0, batch_size):
demo_path = '%s/%s/%s/%s/task_%d/demo_%d' % (
FLAGS.data_path, color_num[color_list[element]], object_num[object_list[element]], FLAGS.demo_type,
task_list[element], demo_list[element])
target_path = '%s/%s/%s/%s/task_%d/demo_%d' % (
FLAGS.data_path, color_num[color_list[element]], object_num[object_list[element]], FLAGS.target_type,
task_list[element], target_list[element])
compare_path = '%s/%s/%s/%s/task_%d/demo_%d' % (
FLAGS.data_path, color_num[compare_color_list[element]], object_num[compare_object_list[element]],
FLAGS.compare_type, compare_task_list[element], compare_demo_list[element])
if FLAGS.cross_domain:
extra_path = '%s/%s/%s/%s/task_%d/demo_%d' % (FLAGS.data_path, color_num[color_list[element]], object_num[object_list[element]],
FLAGS.extra_type, task_list[element], demo_list[element])
if element==0:
if FLAGS.cross_domain:
print('extra_path', extra_path)
print('demo_path', demo_path)
print('target_path', target_path)
print('compare_path', compare_path)
index = np.random.randint(0, FLAGS.index_train_range)
# if if_train:
# index = np.random.randint(0, FLAGS.index_train_range)
# else:
# index = np.random.randint(FLAGS.index_train_range, FLAGS.index_range)
if FLAGS.demo_type == 'robot':
obsa, statea, actiona = read_data.Read_Robot_Data2(demo_path, FLAGS.T, index)
else :
obsa, statea, actiona = read_data.Read_Human_Data2(demo_path, FLAGS.T, index)
if FLAGS.compare_type == 'robot':
obsc, statec, actionc = read_data.Read_Robot_Data2(compare_path, FLAGS.T, index)
else:
obsc, statec, actionc = read_data.Read_Human_Data2(compare_path, FLAGS.T, index)
if FLAGS.cross_domain:
if FLAGS.extra_type == 'robot':
extra_obs, extra_state, extra_action = read_data.Read_Robot_Data2(extra_path, FLAGS.T, index)
else:
extra_obs, extra_state, extra_action = read_data.Read_Human_Data2(extra_path, FLAGS.T, index)
obsb, stateb, actionb = read_data.Read_Robot_Data2(target_path, FLAGS.T, index)
if FLAGS.tar_mil:
labelab = to_categorical(color_list[element] * FLAGS.index_train_range + index, FLAGS.color_num*FLAGS.index_train_range)
labelc = to_categorical(compare_color_list[element] * FLAGS.index_train_range + index, FLAGS.color_num*FLAGS.index_train_range)
labelabs.append(labelab)
labelcs.append(labelc)
# print('element', element, 'labela and labelb', labelab, 'labelc', labelc)
# print('----------------------------------------------------------------')
obsas.append(obsa)
obsbs.append(obsb)
obscs.append(obsc)
stateas.append(statea)
statebs.append(stateb)
statecs.append(statec)
actionas.append(actiona)
actionbs.append(actionb)
actioncs.append(actionc)
if FLAGS.cross_domain:
extra_obses.append(extra_obs)
extra_states.append(extra_state)
extra_actions.append(extra_action)
obsas = np.reshape(obsas, [batch_size, FLAGS.T, FLAGS.im_width * FLAGS.im_height * FLAGS.num_channels])
obsbs = np.reshape(obsbs, [batch_size, FLAGS.T, FLAGS.im_width * FLAGS.im_height * FLAGS.num_channels])
obscs = np.reshape(obscs, [batch_size, FLAGS.T, FLAGS.im_width * FLAGS.im_height * FLAGS.num_channels])
actionas = np.reshape(actionas, [batch_size, FLAGS.T, FLAGS.output_data])
actionbs = np.reshape(actionbs, [batch_size, FLAGS.T, FLAGS.output_data])
actioncs = np.reshape(actioncs, [batch_size, FLAGS.T, FLAGS.output_data])
stateas = np.zeros([batch_size, FLAGS.T, FLAGS.output_data])
statebs = np.zeros([batch_size, FLAGS.T, FLAGS.output_data])
statecs = np.zeros([batch_size, FLAGS.T, FLAGS.output_data])
if FLAGS.tar_mil:
labelabs = np.reshape(labelabs, [batch_size, 1, -1])
labelcs = np.reshape(labelcs, [batch_size, 1, -1])
# print('labelabs', labelabs.shape, 'labelcs', labelcs)
else:
labelabs = np.zeros([batch_size, 1, FLAGS.color_num*FLAGS.index_train_range])
labelcs = np.zeros( [batch_size, 1, FLAGS.color_num*FLAGS.index_train_range])
if FLAGS.cross_domain:
extra_obses = np.reshape(extra_obses, [batch_size, FLAGS.T, FLAGS.im_width * FLAGS.im_height * FLAGS.num_channels])
extra_actions = np.reshape(extra_actions, [batch_size, FLAGS.T, FLAGS.output_data])
extra_states = np.zeros([batch_size, FLAGS.T, FLAGS.output_data])
return obsas, obsbs, obscs, extra_obses, \
actionas, actionbs, actioncs, extra_actions, \
stateas, statebs, statecs, extra_states, \
labelabs, labelcs
else:
return obsas, obsbs, obscs, actionas, actionbs, actioncs, stateas, statebs, statecs, labelabs, labelcs