def run(self, feeds=None):
"""
Run tensors and dump to summary file.
"""
self.make()
result = ThisSession.run(self.summary_op, feeds)
self.file_writer.add_summary(result, ThisSession.run(GlobalStep()))
def main_basic(job, task):
cfg = {"worker": ["localhost:2222",
"localhost:2223"]}
make_distribute_host(cfg, job, task, None, 'worker', 0)
master_host = Master.master_host()
this_host = ThisHost.host()
host1 = Host(job, 1)
hmi = DistributeGraphInfo(None, None, None, master_host)
with tf.variable_scope('scope_test'):
t0 = TensorVariable(VariableInfo(None, [1], tf.float32),
hmi.update(name='v0'))
aop = tf.assign(t0.data, tf.constant([3.]))
t1 = TensorNumpyNDArray([1.0], None,
hmi.update(name='v1'))
t1c = t1.copy_to(host1)
t1p = Tensor(t1c.data + 1, t1c.data_info,
t1c.graph_info.update(name='t1_plus'))
make_distribute_session()
if task == 0:
ptensor(t1)
Server.join()
if task == 1:
ptensor(t1)
ptensor(t1c)
ptensor(t1p)
ptensor(t0)
ThisSession.run(aop)
ptensor(t0)
def bind_local_data_splitted(self, lors):
step = 10000
nb_osem = 10
for i in range(nb_osem):
self.worker_graphs[self.task_index].tensors['osem_{}'.format(
i)] = self.tensor('lorx').assign(lors[i * step:(i + 1) * step,
...])
# when run
ThisSession.run()
def train(self, name=None, feeds=None):
if name is None:
name = 'default'
trainer = self.tensors.get(name)
if trainer is None:
raise ValueError("Nothing to train, please bind first.")
ThisSession.run(trainer, feeds)
global_step = ThisSession.run(self.global_step.increased())
self.on_step_end(name, global_step)
def run_task(self):
KT = self.KEYS.TENSOR
KC = self.KEYS.CONFIG
ThisSession.run(self.tensors[KT.INIT])
for i in tqdm(range(self.config[KC.NB_ITERATIONS])):
ThisSession.run(self.tensors[KT.RECON])
ThisSession.run(self.tensors[KT.MERGE])
if ThisHost.is_master():
x = ThisSession.run(self.tensors[KT.X].data)
np.save(f'./result_{i}.npy', x)
def main(job, task):
tf.logging.set_verbosity(0)
cfg = {"worker": ["localhost:2222",
"localhost:2223"]}
make_distribute_host(cfg, job, task, None, 'worker', 0)
# # if task == 1:
# # time.sleep(10)
# with tf.device(Master.master_host().device_prefix()):
# with tf.variable_scope('test'):
# t1 = tf.get_variable('var', [], tf.float32)
master_host = Master.master_host()
this_host = ThisHost.host()
host2 = Host(job, 1)
hmi = DistributeGraphInfo(None, None, None, master_host)
with tf.variable_scope('scope_test'):
t0 = TensorVariable(VariableInfo(None, [1], tf.float32),
DistributeGraphInfo.from_(hmi, name='t1'))
aop = tf.assign(t0.data, tf.constant([3.]))
t1 = TensorNumpyNDArray([1.0], None,
DistributeGraphInfo.from_(hmi, name='t1_copy'))
t1c = t1.copy_to(host2)
t1p = Tensor(t1c.data + 1, t1c.data_info, DistributeGraphInfo.from_(t1c.graph_info, name='t1_plus'))
# t2 = t0.copy_to(host2)
make_distribute_session()
if task == 0:
# ThisSession.run(tf.global_variables_initializer())
ptensor(t1)
Server.join()
if task == 1:
ptensor(t1)
ptensor(t1c)
ptensor(t1p)
# print(t2.run())
# print(t2.data)
# print(t0.run())
# print(t0)
ptensor(t0)
print(ThisSession.run(aop))
ptensor(t0)
def test_phase(self):
ThisSession.run(self.assign_to_one)
yield
ThisSession.run(self.assign_to_init)
def run_and_save_if_is_master(self, x, path):
if ThisHost.is_master():
if isinstance(x, Tensor):
x = x.data
result = ThisSession.run(x)
np.save(path, result)
def main_sync(job, task):
cfg = {"master": ["localhost:2221"],
"worker": ["localhost:2222",
"localhost:2223"]}
make_distribute_host(cfg, job, task, None, 'master', 0)
master_host = Master.master_host()
this_host = ThisHost.host()
host0 = Host('worker', 0)
host1 = Host('worker', 1)
def sleep(ips):
for i in range(5, 0, -1):
time.sleep(1)
return 0
# hmi = DistributeGraphInfo(None, None, None, master_host)
tm = TensorNumpyNDArray([1.0], None,
DistributeGraphInfo.from_graph_info(hmi, name='t0'))
tcs = []
# t0c = tm.copy_to(host0)
# t1c = tm.copy_to(host1)
# m_sum = Summation(name='summation', graph_info=DistributeGraphInfo(
# 'summation', None, None, host0))([t0c, t1c])
ops = tf.FIFOQueue(2, tf.bool, shapes=[],
name='barrier', shared_name='barrier')
# ptensor(tm)
if ThisHost.host() == master_host:
join = ops.dequeue_many(2)
else:
signal = ops.enqueue(False)
no = tf.constant('tmp')
ops = [tf.Print(no, data=[no], message='Done_{}'.format(i), name='p_{}'.format(i))
for i in range(3)]
# ops.enqueue()
make_distribute_session()
if ThisHost.host() == master_host:
ThisSession.run(join)
print('Joined.')
time.sleep(2)
ThisSession.run(ops[0])
# Server.join()
elif ThisHost.host() == host0:
ThisSession.run(signal)
ThisSession.run(ops[1])
elif ThisHost.host() == host1:
time.sleep(3)
ThisSession.run(signal)
ThisSession.run(ops[2])
def run(self, feeds=None):
self.make()
result = ThisSession.run(self.summary_op, feeds)
self.file_writer.add_summary(result, ThisSession.run(GlobalStep()))
def auto_run(self, feeds=None):
if ThisSession.run(GlobalStep()) >= self._next_summary_step:
self.run(feeds)
self._next_summary_step += self.config(
self.KEYS.CONFIG.NB_INTERVAL)
def evaluate(self, name=None, feeds=None):
with get_global_context().test_phase():
return ThisSession.run(self.tensors.get(name), feeds)
def run_and_print_if_is_master(self, x):
if ThisHost.is_master():
if isinstance(x, Tensor):
x = x.data
result = ThisSession.run(x)
print(result)