def setUpClass(cls): os.environ['CPU_NUM'] = str(4) cls.word_dict_len = 5147 batch_size = 64 reader = fake_imdb_reader(cls.word_dict_len, batch_size * 100) reader = paddle.batch(reader, batch_size=batch_size)() cls.train_data = next(reader)
def setUp(self): self.word_dict_len = 5147 self.BATCH_SIZE = 2 reader = fake_imdb_reader(self.word_dict_len, self.BATCH_SIZE * 100) self.train_data = paddle.batch(reader, batch_size=self.BATCH_SIZE) self.clip_gradient = lambda x: None self.init()
def train(network, use_cuda, use_parallel_executor, batch_size=32, pass_num=2): if use_cuda and not core.is_compiled_with_cuda(): print('Skip use_cuda=True because Paddle is not compiled with cuda') return if use_parallel_executor and os.name == 'nt': print( 'Skip use_parallel_executor=True because Paddle comes without parallel support on windows' ) return word_dict_size = 5147 reader = fake_imdb_reader(word_dict_size, batch_size * 40) train_reader = paddle.batch(reader, batch_size=batch_size) data = fluid.layers.data(name="words", shape=[1], dtype="int64", lod_level=1) label = fluid.layers.data(name="label", shape=[1], dtype="int64") cost = network(data, label, word_dict_size) cost.persistable = True optimizer = fluid.optimizer.Adagrad(learning_rate=0.2) optimizer.minimize(cost) place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace() feeder = fluid.DataFeeder(feed_list=[data, label], place=place) reader = feeder.decorate_reader(train_reader, multi_devices=use_parallel_executor) exe = fluid.Executor(place) fluid.default_startup_program().random_seed = 1 fluid.default_main_program().random_seed = 1 exe.run(fluid.default_startup_program()) train_cp = fluid.default_main_program() if use_parallel_executor: train_cp = compiler.CompiledProgram( fluid.default_main_program()).with_data_parallel( loss_name=cost.name) fetch_list = [cost.name] else: fetch_list = [cost] for pass_id in six.moves.xrange(pass_num): batch_id = 0 for data in reader(): exe.run(train_cp, feed=data, fetch_list=fetch_list if batch_id % 4 == 0 else []) batch_id += 1 if batch_id > 16: break
def test_lstm(self): def optimizer(): optimizer = fluid.optimizer.Adagrad( learning_rate=0.001, regularization=fluid.regularizer.L2Decay(1e-4)) return optimizer with self.program_scope_guard(): word_dict_size = 5147 reader = fake_imdb_reader(word_dict_size, 1) data = fluid.layers.data( name="words", shape=[1], dtype="int64", lod_level=1) label = fluid.layers.data(name="label", shape=[1], dtype="int64") feeder = fluid.DataFeeder( feed_list=[data, label], place=core.CPUPlace()) feed_data = feeder.feed(reader()) self.check_prune_correctness( method=lstm_net, feed_dict=feed_data, optimizer=optimizer)
def setUp(self): self.word_dict_len = 5147 self.BATCH_SIZE = 2 reader = fake_imdb_reader(self.word_dict_len, self.BATCH_SIZE * 100) self.train_data = paddle.batch(reader, batch_size=self.BATCH_SIZE)