def train(conf_dict, to_static):
"""
train process
"""
program_translator = ProgramTranslator()
program_translator.enable(to_static)
# Get device
if fluid.is_compiled_with_cuda():
place = fluid.CUDAPlace(0)
else:
place = fluid.CPUPlace()
with fluid.dygraph.guard(place):
paddle.manual_seed(SEED)
paddle.framework.random._manual_program_seed(SEED)
conf_dict['dict_size'] = len(vocab)
conf_dict['seq_len'] = args.seq_len
net = BOW(conf_dict)
loss = HingeLoss(conf_dict)
optimizer = fluid.optimizer.AdamOptimizer(
learning_rate=0.001,
beta1=0.9,
beta2=0.999,
epsilon=1e-08,
parameter_list=net.parameters())
metric = fluid.metrics.Auc(name="auc")
global_step = 0
losses = []
train_loader = fluid.io.DataLoader.from_generator(
capacity=16,
return_list=True,
iterable=True,
use_double_buffer=True)
get_train_examples = simnet_process.get_reader("train",
epoch=args.epoch)
train_loader.set_sample_list_generator(
paddle.batch(get_train_examples, batch_size=args.batch_size),
place)
for left, pos_right, neg_right in train_loader():
left = fluid.layers.reshape(left, shape=[-1, 1])
pos_right = fluid.layers.reshape(pos_right, shape=[-1, 1])
neg_right = fluid.layers.reshape(neg_right, shape=[-1, 1])
net.train()
global_step += 1
left_feat, pos_score = net(left, pos_right)
pred = pos_score
_, neg_score = net(left, neg_right)
avg_cost = loss.compute(pos_score, neg_score)
losses.append(np.mean(avg_cost.numpy()))
avg_cost.backward()
optimizer.minimize(avg_cost)
net.clear_gradients()
return losses
class TestReturnBase(unittest.TestCase):
def setUp(self):
self.input = np.ones((1)).astype('int32')
self.place = fluid.CUDAPlace(
0) if fluid.is_compiled_with_cuda() else fluid.CPUPlace()
self.init_dygraph_func()
self.program_translator = ProgramTranslator()
def init_dygraph_func(self):
self.dygraph_func = test_return_base
def run_dygraph_mode(self):
self.program_translator.enable(False)
with fluid.dygraph.guard():
res = self.dygraph_func(self.input)
if isinstance(res, (tuple)):
return tuple(r.numpy() for r in res)
elif isinstance(res, core.VarBase):
return res.numpy()
return res
def run_static_mode(self):
self.program_translator.enable(True)
with fluid.dygraph.guard():
res = self.dygraph_func(self.input)
if isinstance(res, tuple):
return tuple(r.numpy() for r in res)
elif isinstance(res, core.VarBase):
return res.numpy()
return res
def test_transformed_static_result(self):
dygraph_res = self.run_dygraph_mode()
static_res = self.run_static_mode()
if isinstance(dygraph_res, tuple):
self.assertTrue(isinstance(static_res, tuple))
self.assertEqual(len(dygraph_res), len(static_res))
for i in range(len(dygraph_res)):
self.assertTrue(
np.allclose(dygraph_res[i], static_res[i]),
msg='dygraph res is {}\nstatic_res is {}'.format(
dygraph_res[i], static_res[i]))
elif isinstance(dygraph_res, np.ndarray):
self.assertTrue(np.allclose(dygraph_res, static_res),
msg='dygraph res is {}\nstatic_res is {}'.format(
dygraph_res, static_res))
else:
self.assertEqual(dygraph_res, static_res)
def predict_dygraph(data):
program_translator = ProgramTranslator()
program_translator.enable(False)
with fluid.dygraph.guard(place):
se_resnext = SeResNeXt()
model_dict, _ = fluid.dygraph.load_dygraph(DY_STATE_DICT_SAVE_PATH)
se_resnext.set_dict(model_dict)
se_resnext.eval()
label = np.random.random([1, 1]).astype("int64")
img = fluid.dygraph.to_variable(data)
label = fluid.dygraph.to_variable(label)
pred_res, _, _, _ = se_resnext(img, label)
return pred_res.numpy()
def test_switch_eval_and_train(self):
program_translator = ProgramTranslator()
with fluid.dygraph.guard():
linear_net = Linear()
x_data = np.random.random((4, 10)).astype('float32')
x = fluid.dygraph.to_variable(x_data)
linear_net(x)
_, partial_layer = linear_net.forward.program_cache.last()[-1]
# check default mode is for training
self.assertEqual(partial_layer.program,
partial_layer._train_program)
# switch to run test program after `eval()`
linear_net.eval()
linear_net(x)
self.assertEqual(partial_layer.program,
partial_layer._infer_program)
# switch back into training
linear_net.train()
linear_net(x)
self.assertEqual(partial_layer.program,
partial_layer._train_program)
def train(to_static):
program_translator = ProgramTranslator()
program_translator.enable(to_static)
random.seed(0)
np.random.seed(0)
place = fluid.CUDAPlace(0) if fluid.is_compiled_with_cuda(
) else fluid.CPUPlace()
with fluid.dygraph.guard(place):
fluid.default_startup_program().random_seed = 1000
fluid.default_main_program().random_seed = 1000
skip_gram_model = SkipGram("skip_gram_model", vocab_size,
embedding_size)
adam = fluid.optimizer.AdamOptimizer(
learning_rate=learning_rate,
parameter_list=skip_gram_model.parameters())
step = 0
ret = []
for center_words, target_words, label, eval_words in build_batch(
dataset, batch_size, epoch_num):
center_words_var = fluid.dygraph.to_variable(center_words)
target_words_var = fluid.dygraph.to_variable(target_words)
label_var = fluid.dygraph.to_variable(label)
pred, loss = skip_gram_model(center_words_var, target_words_var,
label_var)
loss.backward()
adam.minimize(loss)
skip_gram_model.clear_gradients()
step += 1
mean_loss = np.mean(loss.numpy())
print("step %d / %d, loss %f" % (step, total_steps, mean_loss))
ret.append(mean_loss)
return np.array(ret)
def test_save_load(self):
# train and save model
train_layer = self.train_and_save_model()
# load model
program_translator = ProgramTranslator()
program_translator.enable(False)
loaded_layer = fluid.dygraph.jit.load(self.model_path)
self.load_and_inference(train_layer, loaded_layer)
self.load_dygraph_state_dict(train_layer)
self.load_and_finetune(train_layer, loaded_layer)
program_translator.enable(True)
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import numpy as np
import unittest
import paddle
import paddle.fluid as fluid
from paddle.static import InputSpec
from paddle.fluid.dygraph import to_variable, declarative, ProgramTranslator, Layer, jit
from paddle.fluid.dygraph.dygraph_to_static.program_translator import ConcreteProgram, StaticFunction
from test_basic_api_transformation import dyfunc_to_variable
program_trans = ProgramTranslator()
class SimpleNet(Layer):
def __init__(self):
super(SimpleNet, self).__init__()
self.linear = fluid.dygraph.Linear(10, 3)
@declarative(input_spec=[InputSpec(shape=[None, 10], dtype='float32')])
def forward(self, x, a=1, b=2):
y = self.inner_function(x)
return y
# `declarative` is not essential, add it to test for robustness.
@declarative
def inner_function(self, x):
def train(to_static):
program_translator = ProgramTranslator()
program_translator.enable(to_static)
random.seed(0)
np.random.seed(0)
place = fluid.CUDAPlace(0) if cfg.use_gpu else fluid.CPUPlace()
with fluid.dygraph.guard(place):
fluid.default_startup_program().random_seed = 1000
fluid.default_main_program().random_seed = 1000
model = YOLOv3(3, is_train=True)
boundaries = cfg.lr_steps
gamma = cfg.lr_gamma
step_num = len(cfg.lr_steps)
learning_rate = cfg.learning_rate
values = [learning_rate * (gamma**i) for i in range(step_num + 1)]
lr = fluid.dygraph.PiecewiseDecay(boundaries=boundaries,
values=values,
begin=0)
lr = fluid.layers.linear_lr_warmup(
learning_rate=lr,
warmup_steps=cfg.warm_up_iter,
start_lr=0.0,
end_lr=cfg.learning_rate,
)
optimizer = fluid.optimizer.Momentum(
learning_rate=lr,
regularization=fluid.regularizer.L2Decay(cfg.weight_decay),
momentum=cfg.momentum,
parameter_list=model.parameters())
start_time = time.time()
snapshot_loss = 0
snapshot_time = 0
total_sample = 0
input_size = cfg.input_size
shuffle = True
shuffle_seed = None
total_iter = cfg.max_iter
mixup_iter = total_iter - cfg.no_mixup_iter
train_reader = FakeDataReader().reader()
smoothed_loss = SmoothedValue()
ret = []
for iter_id, data in enumerate(train_reader()):
prev_start_time = start_time
start_time = time.time()
img = np.array([x[0] for x in data]).astype('float32')
img = to_variable(img)
gt_box = np.array([x[1] for x in data]).astype('float32')
gt_box = to_variable(gt_box)
gt_label = np.array([x[2] for x in data]).astype('int32')
gt_label = to_variable(gt_label)
gt_score = np.array([x[3] for x in data]).astype('float32')
gt_score = to_variable(gt_score)
loss = model(img, gt_box, gt_label, gt_score, None, None)
smoothed_loss.add_value(np.mean(loss.numpy()))
snapshot_loss += loss.numpy()
snapshot_time += start_time - prev_start_time
total_sample += 1
print("Iter {:d}, loss {:.6f}, time {:.5f}".format(
iter_id, smoothed_loss.get_mean_value(),
start_time - prev_start_time))
ret.append(smoothed_loss.get_mean_value())
loss.backward()
optimizer.minimize(loss)
model.clear_gradients()
return np.array(ret)
def train(args, fake_data_reader, to_static):
program_translator = ProgramTranslator()
program_translator.enable(to_static)
config = parse_config(args.config)
train_config = merge_configs(config, 'train', vars(args))
valid_config = merge_configs(config, 'valid', vars(args))
print_configs(train_config, 'Train')
place = fluid.CUDAPlace(0) if args.use_gpu else fluid.CPUPlace()
random.seed(0)
np.random.seed(0)
with fluid.dygraph.guard(place):
paddle.seed(1000)
paddle.framework.random._manual_program_seed(1000)
video_model = TSM_ResNet("TSM", train_config, 'Train')
optimizer = create_optimizer(train_config.TRAIN,
video_model.parameters())
train_reader = fake_data_reader.create_reader()
ret = []
for epoch in range(train_config.TRAIN.epoch):
video_model.train()
total_loss = 0.0
total_acc1 = 0.0
total_acc5 = 0.0
total_sample = 0
for batch_id, data in enumerate(train_reader()):
x_data = np.array([item[0] for item in data])
y_data = np.array([item[1] for item in data]).reshape([-1, 1])
imgs = to_variable(x_data)
labels = to_variable(y_data)
labels.stop_gradient = True
outputs = video_model(imgs)
loss = fluid.layers.cross_entropy(input=outputs,
label=labels,
ignore_index=-1)
avg_loss = fluid.layers.mean(loss)
acc_top1 = fluid.layers.accuracy(input=outputs,
label=labels,
k=1)
acc_top5 = fluid.layers.accuracy(input=outputs,
label=labels,
k=5)
avg_loss.backward()
optimizer.minimize(avg_loss)
video_model.clear_gradients()
total_loss += avg_loss.numpy()[0]
total_acc1 += acc_top1.numpy()[0]
total_acc5 += acc_top5.numpy()[0]
total_sample += 1
print('TRAIN Epoch {}, iter {}, loss = {}, acc1 {}, acc5 {}'.
format(epoch, batch_id,
avg_loss.numpy()[0],
acc_top1.numpy()[0],
acc_top5.numpy()[0]))
ret.extend([
avg_loss.numpy()[0],
acc_top1.numpy()[0],
acc_top5.numpy()[0]
])
print(
'TRAIN End, Epoch {}, avg_loss= {}, avg_acc1= {}, avg_acc5= {}'
.format(epoch, total_loss / total_sample,
total_acc1 / total_sample, total_acc5 / total_sample))
return ret
def train(train_reader, to_static):
program_translator = ProgramTranslator()
program_translator.enable(to_static)
np.random.seed(SEED)
with fluid.dygraph.guard(place):
paddle.seed(SEED)
paddle.framework.random._manual_program_seed(SEED)
se_resnext = SeResNeXt()
optimizer = optimizer_setting(train_parameters,
se_resnext.parameters())
for epoch_id in range(EPOCH_NUM):
total_loss = 0.0
total_acc1 = 0.0
total_acc5 = 0.0
total_sample = 0
step_idx = 0
speed_list = []
for step_id, data in enumerate(train_reader()):
dy_x_data = np.array([x[0].reshape(3, 224, 224)
for x in data]).astype('float32')
y_data = np.array([x[1] for x in data
]).astype('int64').reshape(BATCH_SIZE, 1)
img = to_variable(dy_x_data)
label = to_variable(y_data)
label.stop_gradient = True
pred, avg_loss, acc_top1, acc_top5 = se_resnext(img, label)
dy_out = avg_loss.numpy()
avg_loss.backward()
optimizer.minimize(avg_loss)
se_resnext.clear_gradients()
lr = optimizer._global_learning_rate().numpy()
total_loss += dy_out
total_acc1 += acc_top1.numpy()
total_acc5 += acc_top5.numpy()
total_sample += 1
if step_id % PRINT_STEP == 0:
if step_id == 0:
logging.info( "epoch %d | step %d, loss %0.3f, acc1 %0.3f, acc5 %0.3f" % \
( epoch_id, step_id, total_loss / total_sample, \
total_acc1 / total_sample, total_acc5 / total_sample))
avg_batch_time = time.time()
else:
speed = PRINT_STEP / (time.time() - avg_batch_time)
speed_list.append(speed)
logging.info( "epoch %d | step %d, loss %0.3f, acc1 %0.3f, acc5 %0.3f, speed %.3f steps/s" % \
( epoch_id, step_id, total_loss / total_sample, \
total_acc1 / total_sample, total_acc5 / total_sample, speed))
avg_batch_time = time.time()
step_idx += 1
if step_idx == STEP_NUM:
if to_static:
fluid.dygraph.jit.save(se_resnext,
MODEL_SAVE_PREFIX, [img],
output_spec=[pred])
else:
fluid.dygraph.save_dygraph(se_resnext.state_dict(),
DY_STATE_DICT_SAVE_PATH)
break
return pred.numpy(), avg_loss.numpy(), acc_top1.numpy(
), acc_top5.numpy()
def setUp(self):
self.input = np.ones((1)).astype('int32')
self.place = fluid.CUDAPlace(
0) if fluid.is_compiled_with_cuda() else fluid.CPUPlace()
self.init_dygraph_func()
self.program_translator = ProgramTranslator()