def measure_denoising(n_datasets, ip_file_path):
data_ids = range(n_datasets)
ip_ports, ips = dh.read_ip_file(ip_file_path)
tf_cluster = dt.TfCluster(ip_ports[:])
datasets = []
masks = []
for data_id in data_ids:
data, bvals_path, bvecs_path = dh.download(data_id)
datasets.append(data)
gtab = dpg.gradient_table(bvals_path, bvecs_path, b0_threshold=10)
mean_b0 = np.mean(data[..., gtab.b0s_mask], -1)
_, mask = median_otsu(mean_b0,
4,
2,
False,
vol_idx=np.where(gtab.b0s_mask),
dilate=1)
masks.append(mask)
with tf.Session("grpc://%s" % tf_cluster.host) as sess:
dt.parallel_denoise(sess, tf_cluster, datasets, masks, depth=1)
def measure_mask(n_datasets, ip_file_path):
data_ids = range(n_datasets)
ip_ports, ips = dh.read_ip_file(ip_file_path)
tf_cluster = dt.TfCluster(ip_ports[:])
filtered_data = []
for data_id in data_ids:
data, bvals_path, bvecs_path = dh.download(data_id)
gtab = dpg.gradient_table(bvals_path, bvecs_path, b0_threshold=10)
filtered_data.append(data[..., gtab.b0s_mask])
with tf.Session("grpc://%s" % tf_cluster.host) as sess:
print("\nMean\n")
dt.parallel_mean(sess, tf_cluster, filtered_data)
def end_to_end(n_datasets, ip_file_path, stride=8):
time_total_start = time.time()
data_ids = range(n_datasets)
ip_ports, ips = dh.read_ip_file(ip_file_path)
tf_cluster = dt.TfCluster(ip_ports[:])
for data_ids_block in [
data_ids[i:i + stride] for i in range(0, len(data_ids), stride)
]:
with tf.Session("grpc://%s" % tf_cluster.host) as sess:
time_iteration = time.time()
end_to_end_iteration(sess, tf_cluster, data_ids_block)
print("Time iteration: %.3fs" % (time.time() - time_iteration))
tf.reset_default_graph()
print("Time overall: %.3fs" % (time.time() - time_total_start))
def measure_model_building(n_datasets, ip_file_path, stride=9):
all_data_ids = range(n_datasets)
for data_ids in [
all_data_ids[i:i + stride]
for i in range(0, len(all_data_ids), stride)
]:
ip_ports, ips = dh.read_ip_file(ip_file_path)
tf_cluster = dt.TfCluster(ip_ports[:])
datasets = []
masks = []
gtabs = []
for data_id in data_ids:
data, bvals_path, bvecs_path = dh.download(data_id)
datasets.append(data)
gtab = dpg.gradient_table(bvals_path, bvecs_path, b0_threshold=10)
mean_b0 = np.mean(data[..., gtab.b0s_mask], -1)
_, mask = median_otsu(mean_b0,
4,
2,
False,
vol_idx=np.where(gtab.b0s_mask),
dilate=1)
masks.append(mask)
gtabs.append(gtab)
print("Start measurement")
time_start = time.time()
with tf.Session("grpc://%s" % tf_cluster.host) as sess:
dt.parallel_modelbuilding(sess,
tf_cluster,
masks,
datasets,
gtabs,
n_parts=64)
print("Time iteration: %.3fs" % (time.time() - time_start))
def measure_image_filtering(n_datasets, multiplicator, ip_file_path):
data_ids = range(n_datasets)
ip_ports, ips = dh.read_ip_file(ip_file_path)
tf_cluster = dt.TfCluster(ip_ports)
datasets = []
path_sets = []
for data_id in data_ids:
data, bvals_path, bvecs_path = dh.download(data_id)
datasets.append(data)
print(datasets[-1].shape)
path_sets.append([bvals_path, bvecs_path])
with tf.Session("grpc://%s" % tf_cluster.host) as sess:
print("\nImage filtering\n")
dt.parallel_image_filtering(sess,
tf_cluster,
datasets,
path_sets,
multiplicator,
n_parts=288)