def test_filler_to_malloc(self): from adlkit.data_provider.tests.mock_config import mock_sample_specification, \ mock_expected_malloc_requests mock_sample_specification = copy.deepcopy(mock_sample_specification) mock_expected_malloc_requests = copy.deepcopy( mock_expected_malloc_requests) batch_size = 100 tmp_data_provider = FileDataProvider(mock_sample_specification, batch_size=batch_size, n_readers=4, sleep_duration=sleep_duration) tmp_data_provider.start_queues() tmp_data_provider.start_filler(FileFiller, io_ctlr=IOController(), shape_reader_io_ctlr=None) tmp_data_provider.process_malloc_requests() tmp_data_provider.worker_count = worker_id = 1 tmp_data_provider.make_shared_malloc() bucket = 0 data_set = 0 lock = 0 # TODO !!possible off by one error, needs logic check!! # check to make sure the shared mem was extended # self.assertEqual(len(tmp_data_provider.shared_memory), # tmp_data_provider.worker_count + 1, # "shared memory was not extended correctly ") # check to make sure all buckets were allocated self.assertEqual(len(tmp_data_provider.shared_memory), 10, "shared memory buckets were not allocated correctly") # Multiple Generator start and end locks self.assertEqual(len(tmp_data_provider.shared_memory[bucket]), 4, "shared memory locks were not set correctly") # check to make sure all data sets were allocated self.assertEqual( len(tmp_data_provider.shared_memory[bucket][lock + 1]), len(mock_expected_malloc_requests), "shared memory data sets were not allocated correctly") # check to make sure the shape matches out expected value self.assertEqual( tmp_data_provider.shared_memory[bucket][lock + 1][data_set].shape, (100, 5), "shared memory shape doesn't match") tmp_data_provider.hard_stop()
def test_class_index_map(self): """ if this fails, it's most likely due to a lack of allowed file descriptors :return: """ from adlkit.data_provider.tests.mock_config import mock_sample_specification mock_sample_specification = copy.deepcopy(mock_sample_specification) batch_size = 5 tmp_data_provider = FileDataProvider(mock_sample_specification, batch_size=batch_size, read_batches_per_epoch=1000, read_multiplier=1, make_one_hot=True, make_class_index=True, wrap_examples=True, n_readers=5, n_buckets=2, q_multipler=3, sleep_duration=sleep_duration, class_index_map={ 'class_1': -3, 'class_2': -2, 'class_10': -1 }) tmp_data_provider.start(filler_class=FileFiller, reader_class=FileReader, generator_class=BaseGenerator) generator_id = 0 for _ in range(100): # for _ in range(10000000): tmp = tmp_data_provider.generators[generator_id].generate().next() # TODO better checks self.assertEqual(len(tmp), 4) tmp_data_provider.hard_stop() # we are checking the that readers are killing themselves # by checking that the pid no longer exists # https://stackoverflow.com/questions/568271/how-to-check-if-there-exists-a-process-with-a-given-pid-in-python for reader_process in tmp_data_provider.readers: this = None try: os.kill(reader_process.pid, 0) this = True except OSError: this = False finally: self.assertEqual(this, False)
def test_multiple_generators(self): from adlkit.data_provider.tests.mock_config import mock_sample_specification mock_sample_specification = copy.deepcopy(mock_sample_specification) max_batches = 10 batch_size = 100 n_generators = 5 tmp_data_provider = FileDataProvider(mock_sample_specification, batch_size=batch_size, read_multiplier=2, make_one_hot=True, make_class_index=True, n_readers=3, n_generators=n_generators, wrap_examples=True, sleep_duration=sleep_duration) tmp_data_provider.start(filler_class=FileFiller, reader_class=FileReader, generator_class=BaseGenerator, watcher_class=BaseWatcher) # ugly, but mostly necessary generator_counter = [[] for _ in range(n_generators)] total_count = 0 for loop in range(max_batches): for gen_index in range(len(tmp_data_provider.generators)): this = None this = tmp_data_provider.generators[gen_index].generate().next( ) self.assertEqual(len(this), 4) generator_counter[gen_index].append(this) total_count += 1 for generator_count in generator_counter: self.assertEqual(len(generator_count), max_batches) for p_index, payload in enumerate(generator_counter[0]): for e_index, entry in enumerate(payload): for generator_output in generator_counter: self.assertTrue( np.array_equal(generator_output[p_index][e_index], entry)) self.assertEqual(total_count, max_batches * n_generators) tmp_data_provider.hard_stop()
def test_start_filler(self): from adlkit.data_provider.tests.mock_config import mock_sample_specification mock_sample_specification = copy.deepcopy(mock_sample_specification) tmp_data_provider = FileDataProvider(mock_sample_specification, batch_size=100, read_multiplier=2, n_readers=4, sleep_duration=sleep_duration) tmp_data_provider.start_queues() tmp_data_provider.start_filler(FileFiller, io_ctlr=IOController()) max_batches = 10 out = list() start_time = datetime.datetime.utcnow() end_time = datetime.timedelta(seconds=10) + start_time while datetime.datetime.utcnow() < end_time and len( out) != max_batches: batch = tmp_data_provider.comm_driver.read('in', block=False) if batch is not None: out.append(batch) count = 0 for item in batch: count += item[3][1] - item[3][0] self.assertEquals( count, tmp_data_provider.config.read_size, "batch was returned with batch_size {0}, instead of batch_size {1}" .format(count, tmp_data_provider.config.batch_size)) self.assertEquals( len(out), max_batches, "test consumed {0} of {1} expected batches from the in_queue". format(len(out), max_batches)) tmp_data_provider.hard_stop()
def test_watcher_multicast(self): from adlkit.data_provider.tests.mock_config import mock_read_batches, \ mock_sample_specification, mock_expected_malloc_requests mock_read_batches = copy.deepcopy(mock_read_batches) mock_sample_specification = copy.deepcopy(mock_sample_specification) mock_expected_malloc_requests = copy.deepcopy( mock_expected_malloc_requests) batch_size = 100 max_batches = 5 n_generators = 5 n_readers = 5 tmp_data_provider = FileDataProvider(mock_sample_specification, batch_size=batch_size, read_multipler=2, make_one_hot=True, make_class_index=True, n_readers=n_readers, n_generators=n_generators, n_buckets=2, sleep_duration=sleep_duration) tmp_data_provider.start_queues() tmp_data_provider.malloc_requests = mock_expected_malloc_requests tmp_data_provider.make_shared_malloc() for batch in mock_read_batches: # try: # tmp_data_provider.out_queue.put(batch, timeout=0) success = tmp_data_provider.comm_driver.write('out', batch, block=False) # except Queue.Full: if not success: max_batches -= 1 tmp_watcher = BaseWatcher( worker_id=0, comm_driver=tmp_data_provider.comm_driver, proxy_comm_drivers=tmp_data_provider.proxy_comm_drivers, shared_memory_pointer=tmp_data_provider.shared_memory, # multicast_queues=tmp_data_provider.multicast_queues, max_batches=max_batches, # out_queue=tmp_data_provider.out_queue ) tmp_watcher.watch() gen_counter = [0] * 5 for loop in range(max_batches): for index, comm_driver in enumerate( tmp_data_provider.proxy_comm_drivers): # this = None # try: # this = tmp_data_provider.multicast_queues[gen_index].get() this = comm_driver.read('out', block=False) # except Queue.Empty: # pass # finally: if this is not None: gen_counter[index] += 1 for gen_count in gen_counter: self.assertEqual(gen_count, max_batches) tmp_data_provider.hard_stop()