def start_worker(self):
if singleton_pool.n_parallel > 1:
singleton_pool.initialize(singleton_pool.n_parallel)
singleton_pool.run_each(worker_init_tf)
parrallel_sampler.populate_task(self.algo.env, self.algo.policy)
#if singleton_pool.n_parallel > 1:
# singleton_pool.run_each(worker_init_tf_vars)
print('done')
def train(self):
singleton_pool.initialize(n_parallel=4)
singleton_pool.run_each(_worker_init, [(id, ) for id in range(4)])
singleton_pool.run_each(_worker_populate_task,
[(pickle.dumps(self), None)] * 4)
manager = mp.Manager()
counter = manager.Value('i', 0)
lock = manager.RLock()
results = singleton_pool.run_each(train_thread, [(counter, lock)] * 4)
def initialize(n_parallel):
singleton_pool.initialize(n_parallel)
singleton_pool.run_each(_worker_init,
[(id, )
for id in range(singleton_pool.n_parallel)])
def initialize(n_parallel):
singleton_pool.initialize(n_parallel)
singleton_pool.run_each(
_worker_init, [(id,) for id in xrange(singleton_pool.n_parallel)])
def initialize(n_parallel):
print(("parallel_sampler:initialize n_parallel", n_parallel))
singleton_pool.initialize(n_parallel)
singleton_pool.run_each(_worker_init, [(id,) for id in range(singleton_pool.n_parallel)])
