def _check(d):
a = d['all_steer']
True
b = []
ctr = 0
for e in a:
run_code = e[3]
seg_num = e[0]
offset = e[1]
r = Segment_Data.get_data(run_code,
seg_num,
offset,
3 * P.N_STEPS,
offset + 0,
P.N_FRAMES,
ignore=P.IGNORE,
require_one=P.REQUIRE_ONE,
use_states=P.USE_STATES,
no_images=True)
if r != None:
b.append(e)
if np.mod(ctr, 1000) == 0:
print ctr
ctr += 1
#if ctr > 1000:
# break
return b
def _get_data(d):
run_code = d['run_code']
seg_num = d['seg_num']
offset = d['offset']
True
data = Segment_Data.get_data(run_code,seg_num,offset,P.STRIDE*P.N_STEPS,offset+0,P.N_FRAMES,ignore=P.IGNORE,require_one=P.REQUIRE_ONE,use_states=P.USE_STATES)
return data
def _function_get_data(*args):
Args = args_to_dictionary(args)
True
Data_moment = Segment_Data.get_data(Args[run_code],
Args[seg_num],
Args[offset],
P[STRIDE] * P[N_STEPS],
Args[offset] + 0,
P[N_FRAMES],
ignore=P[IGNORE],
require_one=P[REQUIRE_ONE],
use_states=P[USE_STATES])
return Data_moment
def _get_data_considering_high_low_steer(d):
global ctr_low
global ctr_high
global low_steer
global high_steer
if ctr_low >= len_low_steer:
ctr_low = -1
if ctr_high >= len_high_steer:
ctr_high = -1
if ctr_low == -1:
random.shuffle(low_steer)
ctr_low = 0
if ctr_high == -1:
random.shuffle(high_steer)
ctr_high = 0
if random.random() < 0.5:
choice = low_steer[ctr_low]
ctr_low += 1
else:
choice = high_steer[ctr_high]
ctr_high += 1
run_code = choice[3]
seg_num = choice[0]
offset = choice[1]
data = Segment_Data.get_data(run_code,
seg_num,
offset,
3 * N_STEPS,
offset + 0,
N_FRAMES,
ignore=ignore,
require_one=require_one,
use_states=use_states)
return data
from Parameters_Module import *
exec(identify_file_str)
import data.utils.Segment_Data as Segment_Data
#
hdf5_runs_path = opj(P['BAIR_CAR_DATA_PATH'], 'hdf5/runs')
hdf5_segment_metadata_path = opj(P['BAIR_CAR_DATA_PATH'],
'hdf5/segment_metadata')
Segment_Data.load_Segment_Data(hdf5_segment_metadata_path, hdf5_runs_path)
print('.')
_ = dictionary_access
def Training_Data():
D = {}
True
_(D, dic_type, equals, 'Training_Data')
_(
D, purpose, equals,
d2s(inspect.stack()[0][3], ':',
'Object to hold various data and information for training'))
_(D, train, equals, {})
print('loading train_all_steer...')
#_(D,train,all_data_moment_id_codes, equals, lo(opj(P[BAIR_CAR_DATA_PATH],'train_all_steer')))
D[train][all_data_moment_id_codes] = lo(
opj(P[BAIR_CAR_DATA_PATH], 'train_all_steer'))
_(D, train, ctr, equals, -1)
_(D, train, loss_dic, equals, {})
_(D, val, equals, {})
print('loading val_all_steer...')
def _get_data_considering_high_low_steer_and_valid_trajectory_timestamp(d):
True
global ctr_low
global ctr_high
global low_steer
global high_steer
global counts
global high_loss_key_ctr
global high_loss_keys
if ctr_low >= len_low_steer:
ctr_low = -1
if ctr_high >= len_high_steer:
ctr_high = -1
if ctr_low == -1:
random.shuffle(low_steer)
ctr_low = 0
if ctr_high == -1:
random.shuffle(high_steer)
ctr_high = 0
if random.random() < 0.5:
choice = low_steer[ctr_low]
ctr_low += 1
else:
choice = high_steer[ctr_high]
ctr_high += 1
run_code = choice[3]
seg_num = choice[0]
offset = choice[1]
run_name = Segment_Data.Segment_Data['run_codes'][run_code]
if run_name not in Aruco_Steering_Trajectories.keys():
#print('Run name '+run_name+' not in Aruco_Steering_Trajectories')
return None
if len(Aruco_Steering_Trajectories[run_name].keys()) < 2:
#print('len(Aruco_Steering_Trajectories[run_name].keys()) <= 2')
return None
seg_num_str = str(seg_num)
aruco_matches = []
for i in [0]: #range(N_FRAMES):
timestamp = Segment_Data.Segment_Data['runs'][run_name][
'segments'][seg_num_str]['left_timestamp'][offset + i]
behavioral_mode = np.random.choice([
'Direct_Arena_Potential_Field',
#'Furtive_Arena_Potential_Field',
'Follow_Arena_Potential_Field'
])
#'Play_Arena_Potential_Field'])
desired_direction = np.random.choice([0, 1])
if timestamp in Aruco_Steering_Trajectories[run_name][
behavioral_mode][desired_direction].keys():
aruco_matches.append(timestamp)
if len(aruco_matches) < 1:
return None
data = Segment_Data.get_data(run_code,
seg_num,
offset,
N_STEPS,
offset + 0,
N_FRAMES,
ignore=ignore,
require_one=require_one)
if data != None:
data['behavioral_mode'] = behavioral_mode
data['desired_direction'] = desired_direction
for topic in Aruco_Steering_Trajectories[run_name][
behavioral_mode][desired_direction][timestamp]:
data[topic] = Aruco_Steering_Trajectories[run_name][
behavioral_mode][desired_direction][timestamp][topic]
data['id'] = (run_name, behavioral_mode, desired_direction,
timestamp, run_code, seg_num, offset)
if P.other_cars_only and np.random.random() < 0.85:
if data == None:
return None
if data['other_car_inverse_distances'] == None:
return None
if len(data['other_car_inverse_distances']) == 0:
return None
return data