def test_disjointPath(self):
cfg.k_Nacre= 20
cfg.defaultBackupStrategy= BackupStrategy.TOR_TO_TOR
nacre = Nacre()
nacre.generate()
source = 'h_1_A_1_1'
dest = 'h_3_B_2_1'
paths = []
while True:
path = nacre.findDisjointPath(source,dest,0,paths)
if path is not None:
paths.append(path)
globals.simulatorLogger.info(path.__str__())
else:
break
globals.simulatorLogger.info("Total disjoint paths found %s" % len(paths))
return True
## distributing tasks accross nodes ##
from mpi4py import MPI
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
size = comm.Get_size()
print(rank)
#assuming mp3 for now. TODO: generalize
candidate_files = sorted(data_path.glob('**/*' + feature_name + '.npy'),
key=lambda path: path.parent.__str__())
tasks = distribute_tasks(candidate_files, rank, size)
for i in tasks:
path = candidate_files[i]
feature_file = path.__str__()
if new_feature_name is None:
if keep_feature_name:
new_feature_name = feature_name
else:
new_feature_name = feature_name + "_applied_" + transform_name
base_filename = feature_file[:-(len(feature_name) + 4)]
new_feature_file = base_filename + new_feature_name + ".npy"
if replace_existing or not os.path.isfile(new_feature_file):
features = np.load(feature_file)
transform = pickle.load(
open(
data_path.joinpath(feature_name + '_' + transform_name +
'.pkl'), "rb"))
features = transform.transform(features)
if transform_name == "pca_transform":
def _is_valid_file(path):
try:
path = Path(path.__str__())
return path.exists() and path.is_file()
except TypeError:
return False
def _is_valid_file(path):
try:
path = Path(path.__str__())
return path.exists() and path.is_file()
except (TypeError, OSError, ValueError):
return False
rank = comm.Get_rank()
size = comm.Get_size()
print(rank)
print("creating tensorblocks")
#assuming egg sound format, as used in new BeatSaber format
candidate_audio_files = sorted(data_path.glob('**/*.egg'), key=lambda path: path.parent.__str__())
num_tasks = len(candidate_audio_files)
num_tasks_per_job = num_tasks//size
tasks = list(range(rank*num_tasks_per_job,(rank+1)*num_tasks_per_job))
if rank < num_tasks%size:
tasks.append(size*num_tasks_per_job+rank)
for i in tasks:
path = candidate_audio_files[i]
song_file_path = path.__str__()
# feature files are going to be saved as numpy files
features_file = song_file_path+"_"+feature_name+"_"+str(feature_size)+".npy"
# blocks_reduced_file = song_file_path+"_"+difficulties+"_blocks_reduced_.npy"
blocks_reduced_classes_file = song_file_path+difficulties+"_blocks_reduced_classes_.npy"
level_file_found = False
# find level files with target difficulties that exist
for diff in difficulties.split(","):
if Path(path.parent.__str__()+"/"+diff+".dat").is_file():
level = list(path.parent.glob('./'+diff+'.dat'))[0]
level = level.__str__()
info_file = list(path.parent.glob('./info.dat'))[0]
info_file = info_file.__str__()
level_file_found = True
if not level_file_found:
def get_features(motion_data):
joint_angle_feats = get_rot_matrices_from_axis_angle(
(motion_data['smpl_poses']))
return np.concatenate([joint_angle_feats, motion_data['smpl_trans']], 1)
## distributing tasks accross nodes ##
from mpi4py import MPI
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
size = comm.Get_size()
print(rank)
candidate_motion_files = sorted(data_path.glob('**/*.pkl'),
key=lambda path: path.parent.__str__())
tasks = distribute_tasks(candidate_motion_files, rank, size)
for i in tasks:
path = candidate_motion_files[i]
motion_file_path = path.__str__()
features_file = motion_file_path + "_" + "joint_angles_mats" + ".npy"
if replace_existing or not os.path.isfile(features_file):
motion_data = pickle.load(open(path, "rb"))
features = get_features(motion_data)
print(features.shape)
features = ResampleLinear1D(features, features.shape[0] * 2)
print(features.shape)
np.save(features_file, features)