def test_torch_model_on_batch_begin(self):
lr_scheduler = LRScheduler(model=self.torch_model,
lr_fn=lambda step: fe.schedule.cosine_decay(
step, cycle_length=3750, init_lr=1e-3))
lr_scheduler.system = sample_system_object()
lr_scheduler.system.global_step = 3
lr_scheduler.on_batch_begin(data=self.data)
new_lr = list(self.torch_model.optimizer.param_groups)[0]['lr']
self.assertTrue(math.isclose(new_lr, 0.0009999993, rel_tol=1e-5))
def test_tf_model_on_batch_begin(self):
lr_scheduler = LRScheduler(model=self.tf_model,
lr_fn=lambda step: fe.schedule.cosine_decay(
step, cycle_length=3750, init_lr=1e-3))
lr_scheduler.system = sample_system_object()
lr_scheduler.system.global_step = 3
lr_scheduler.on_batch_begin(data=self.data)
self.assertTrue(
math.isclose(self.tf_model.optimizer.lr.numpy(),
0.0009999973,
rel_tol=1e-5))
def test_torch_model_on_batch_end(self):
model_name = self.torch_model.model_name + '_lr'
lr_scheduler = LRScheduler(model=self.torch_model,
lr_fn=lambda step: fe.schedule.cosine_decay(
step, cycle_length=3750, init_lr=1e-3))
lr_scheduler.system = sample_system_object()
lr_scheduler.system.global_step = 3
lr_scheduler.system.log_steps = 1
lr_scheduler.on_batch_end(data=self.data)
self.assertTrue(
math.isclose(self.data[model_name], 0.001, rel_tol=1e-3))
def _build_estimator(model: Union[tf.keras.Model, torch.nn.Module], trace: Traceability, axis: int = -1):
train_data, eval_data = mnist.load_data()
test_data = eval_data.split(0.5)
batch_size = 32
pipeline = fe.Pipeline(train_data=train_data,
eval_data=eval_data,
test_data=test_data,
batch_size=batch_size,
ops=[ExpandDims(inputs="x", outputs="x", axis=axis), Minmax(inputs="x", outputs="x")])
network = fe.Network(ops=[
ModelOp(model=model, inputs="x", outputs="y_pred"),
CrossEntropy(inputs=("y_pred", "y"), outputs="ce"),
UpdateOp(model=model, loss_name="ce")
])
traces = [
Accuracy(true_key="y", pred_key="y_pred"),
LRScheduler(model=model, lr_fn=lambda step: cosine_decay(step, cycle_length=3750, init_lr=1e-3)),
trace
]
estimator = fe.Estimator(pipeline=pipeline,
network=network,
epochs=1,
traces=traces,
max_train_steps_per_epoch=1,
max_eval_steps_per_epoch=None)
fake_data = tf.ones(shape=(batch_size, 28, 28, 1)) if axis == -1 else torch.ones(size=(batch_size, 1, 28, 28))
model.fe_input_spec = FeInputSpec(fake_data, model)
return estimator
def get_estimator(epochs=2,
batch_size=32,
save_dir=tempfile.mkdtemp()):
# step 1
train_data, eval_data = mnist.load_data()
test_data = eval_data.split(0.5)
pipeline = fe.Pipeline(train_data=train_data,
eval_data=eval_data,
test_data=test_data,
batch_size=batch_size,
ops=[ExpandDims(inputs="x", outputs="x"), Minmax(inputs="x", outputs="x")])
# step 2
model = fe.build(model_fn=LeNet, optimizer_fn="adam")
network = fe.Network(ops=[
ModelOp(model=model, inputs="x", outputs="y_pred"),
CrossEntropy(inputs=("y_pred", "y"), outputs="ce"),
UpdateOp(model=model, loss_name="ce")
])
# step 3
traces = [
Accuracy(true_key="y", pred_key="y_pred"),
BestModelSaver(model=model, save_dir=save_dir, metric="accuracy", save_best_mode="max"),
LRScheduler(model=model, lr_fn=lambda step: cosine_decay(step, cycle_length=3750, init_lr=1e-3))
]
estimator = fe.Estimator(pipeline=pipeline,
network=network,
epochs=epochs,
traces=traces)
return estimator
def get_estimator(epochs=2,
batch_size=32,
max_train_steps_per_epoch=None,
max_eval_steps_per_epoch=None,
save_dir=tempfile.mkdtemp()):
# step 1
train_data, eval_data = mnist.load_data()
test_data = eval_data.split(0.5)
pipeline = fe.Pipeline(train_data=train_data,
eval_data=eval_data,
test_data=test_data,
batch_size=batch_size,
ops=[
ExpandDims(inputs="x", outputs="x"),
Minmax(inputs="x", outputs="x")
],
num_process=0)
# step 2
model = fe.build(model_fn=LeNet, optimizer_fn="adam")
print([f"{idx}: {x.name}" for idx, x in enumerate(model.submodules)])
network = fe.Network(ops=[
Watch(inputs="x"),
ModelOp(model=model,
inputs="x",
outputs=["y_pred", "embedding"],
intermediate_layers='dense'),
CrossEntropy(inputs=("y_pred", "y"), outputs="ce"),
GradientOp(finals="embedding", inputs="x", outputs="grads"),
UpdateOp(model=model, loss_name="ce")
])
# step 3
traces = [
Accuracy(true_key="y", pred_key="y_pred"),
Inspector(),
BestModelSaver(model=model,
save_dir=save_dir,
metric="accuracy",
save_best_mode="max"),
LRScheduler(model=model,
lr_fn=lambda step: cosine_decay(
step, cycle_length=3750, init_lr=1e-3)),
TensorBoard(log_dir="tf_logs",
write_embeddings="embedding",
embedding_labels="y")
]
estimator = fe.Estimator(
pipeline=pipeline,
network=network,
epochs=epochs,
traces=traces,
max_train_steps_per_epoch=max_train_steps_per_epoch,
max_eval_steps_per_epoch=max_eval_steps_per_epoch)
return estimator
def get_estimator(epochs=24, batch_size=128, lr_epochs=100, max_train_steps_per_epoch=None,
save_dir=tempfile.mkdtemp()):
# step 1: prepare dataset
train_data, test_data = load_data()
pipeline = fe.Pipeline(
train_data=train_data,
eval_data=test_data,
batch_size=batch_size,
ops=[
Normalize(inputs="x", outputs="x", mean=(0.4914, 0.4822, 0.4465), std=(0.2471, 0.2435, 0.2616)),
PadIfNeeded(min_height=40, min_width=40, image_in="x", image_out="x", mode="train"),
RandomCrop(32, 32, image_in="x", image_out="x", mode="train"),
Sometimes(HorizontalFlip(image_in="x", image_out="x", mode="train")),
CoarseDropout(inputs="x", outputs="x", mode="train", max_holes=1),
ChannelTranspose(inputs="x", outputs="x"),
Onehot(inputs="y", outputs="y", mode="train", num_classes=10, label_smoothing=0.2)
])
# step 2: prepare network
model = fe.build(model_fn=ResNet9, optimizer_fn="sgd")
network = fe.Network(ops=[
ModelOp(model=model, inputs="x", outputs="y_pred"),
CrossEntropy(inputs=("y_pred", "y"), outputs="ce"),
UpdateOp(model=model, loss_name="ce")
])
# get the max learning rate
lr_max = search_max_lr(pipeline=pipeline, model=model, network=network, epochs=lr_epochs)
lr_min = lr_max / 40
print(f"The maximum LR: {lr_max}, and minimun LR: {lr_min}")
mid_step = int(epochs * 0.45 * len(train_data) / batch_size)
end_step = int(epochs * len(train_data) / batch_size)
# reinitialize the model
model = fe.build(model_fn=ResNet9, optimizer_fn="sgd")
network = fe.Network(ops=[
ModelOp(model=model, inputs="x", outputs="y_pred"),
CrossEntropy(inputs=("y_pred", "y"), outputs="ce"),
UpdateOp(model=model, loss_name="ce")
])
# step 3: prepare estimator
traces = [
Accuracy(true_key="y", pred_key="y_pred"),
BestModelSaver(model=model, save_dir=save_dir, metric="accuracy", save_best_mode="max"),
LRScheduler(model=model, lr_fn=lambda step: super_schedule(step, lr_max, lr_min, mid_step, end_step))
]
estimator = fe.Estimator(pipeline=pipeline,
network=network,
epochs=epochs,
traces=traces,
max_train_steps_per_epoch=max_train_steps_per_epoch)
return estimator
def get_estimator(data_dir=None,
model_dir=tempfile.mkdtemp(),
epochs=20,
em_dim=128,
batch_size=32,
max_train_steps_per_epoch=None,
max_eval_steps_per_epoch=None):
train_ds, eval_ds, test_ds = tednmt.load_data(data_dir, translate_option="pt_to_en")
pt_tokenizer = BertTokenizer.from_pretrained("neuralmind/bert-base-portuguese-cased")
en_tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")
pipeline = fe.Pipeline(
train_data=train_ds,
eval_data=eval_ds,
test_data=test_ds,
batch_size=batch_size,
ops=[
Encode(inputs="source", outputs="source", tokenizer=pt_tokenizer),
Encode(inputs="target", outputs="target", tokenizer=en_tokenizer)
],
pad_value=0)
model = fe.build(
model_fn=lambda: Transformer(num_layers=4,
em_dim=em_dim,
num_heads=8,
ff_dim=512,
input_vocab=pt_tokenizer.vocab_size,
target_vocab=en_tokenizer.vocab_size,
max_pos_enc=1000,
max_pos_dec=1000),
optimizer_fn="adam")
network = fe.Network(ops=[
ShiftData(inputs="target", outputs=("target_inp", "target_real")),
CreateMasks(inputs=("source", "target_inp"),
outputs=("encode_pad_mask", "decode_pad_mask", "dec_look_ahead_mask")),
ModelOp(model=model,
inputs=("source", "target_inp", "encode_pad_mask", "decode_pad_mask", "dec_look_ahead_mask"),
outputs="pred"),
MaskedCrossEntropy(inputs=("pred", "target_real"), outputs="ce"),
UpdateOp(model=model, loss_name="ce")
])
traces = [
MaskedAccuracy(inputs=("pred", "target_real"), outputs="masked_acc", mode="!train"),
BestModelSaver(model=model, save_dir=model_dir, metric="masked_acc", save_best_mode="max"),
LRScheduler(model=model, lr_fn=lambda step: lr_fn(step, em_dim))
]
estimator = fe.Estimator(pipeline=pipeline,
network=network,
traces=traces,
epochs=epochs,
max_train_steps_per_epoch=max_train_steps_per_epoch,
max_eval_steps_per_epoch=max_eval_steps_per_epoch)
return estimator
def get_estimator(epochs=30,
batch_size=128,
seq_length=20,
vocab_size=10000,
data_dir=None,
train_steps_per_epoch=None,
save_dir=tempfile.mkdtemp()):
train_data, eval_data, _, _ = load_data(root_dir=data_dir,
seq_length=seq_length + 1)
pipeline = fe.Pipeline(train_data=train_data,
eval_data=eval_data,
ops=[
CreateInputAndTarget(inputs="x",
outputs=("x", "y")),
Batch(batch_size=batch_size, drop_last=True)
])
# step 2
model = fe.build(
model_fn=lambda: BuildModel(
vocab_size, embedding_dim=300, rnn_units=600),
optimizer_fn=lambda x: torch.optim.SGD(x, lr=1.0, momentum=0.9))
network = fe.Network(ops=[
DimesionAdjust(inputs=("x", "y"), outputs=("x", "y")),
ModelOp(model=model, inputs="x", outputs="y_pred", mode=None),
CrossEntropy(inputs=("y_pred", "y"),
outputs="ce",
form="sparse",
from_logits=True),
UpdateOp(model=model, loss_name="ce")
])
# step 3
traces = [
Perplexity(inputs="ce", outputs="perplexity", mode="eval"),
LRScheduler(model=model,
lr_fn=lambda step: lr_schedule(step, init_lr=1.0)),
BestModelSaver(model=model,
save_dir=save_dir,
metric='perplexity',
save_best_mode='min',
load_best_final=True),
EarlyStopping(monitor="perplexity", patience=5)
]
estimator = fe.Estimator(pipeline=pipeline,
network=network,
epochs=epochs,
traces=traces,
train_steps_per_epoch=train_steps_per_epoch)
return estimator
def search_max_lr(pipeline, model, network, epochs):
traces = [
Accuracy(true_key="y", pred_key="y_pred"), LRScheduler(model=model, lr_fn=lambda step: linear_increase(step))
]
estimator = fe.Estimator(pipeline=pipeline,
network=network,
epochs=epochs,
traces=traces,
max_train_steps_per_epoch=10,
log_steps=10)
print("Running LR range test for super convergence. It will take a while...")
with Suppressor():
summary = estimator.fit("LR_range_test")
best_step = max(summary.history["eval"]["accuracy"].items(), key=lambda k: k[1])[0]
max_lr = summary.history["train"]["model_lr"][best_step]
return max_lr
def get_estimator():
# step 1
train_data, eval_data = mnist.load_data()
pipeline = fe.Pipeline(train_data=train_data,
eval_data=eval_data,
batch_size=32,
ops=[ExpandDims(inputs="x", outputs="x"), Minmax(inputs="x", outputs="x")])
# step 2
model = fe.build(model_fn=LeNet, optimizer_fn="adam")
network = fe.Network(ops=[
ModelOp(model=model, inputs="x", outputs="y_pred"),
CrossEntropy(inputs=("y_pred", "y"), outputs="ce"),
UpdateOp(model=model, loss_name="ce")
])
# step 3
estimator = fe.Estimator(pipeline=pipeline,
network=network,
traces=LRScheduler(model=model, lr_fn="arc"),
epochs=30,
max_train_steps_per_epoch=500)
return estimator
def create_estimator_for_arc(self, model, use_eval, axis):
train_data, eval_data = mnist.load_data()
pipeline = fe.Pipeline(train_data=train_data,
eval_data=eval_data if use_eval else None,
batch_size=8,
ops=[
ExpandDims(inputs="x",
outputs="x",
axis=axis),
Minmax(inputs="x", outputs="x")
])
network = fe.Network(ops=[
ModelOp(model=model, inputs="x", outputs="y_pred"),
CrossEntropy(inputs=("y_pred", "y"), outputs="ce"),
UpdateOp(model=model, loss_name="ce")
])
estimator = fe.Estimator(pipeline=pipeline,
network=network,
epochs=2,
traces=LRScheduler(model=model, lr_fn=ARC(1)),
max_train_steps_per_epoch=10)
return estimator
def get_estimator(epochs=24, batch_size=512, max_train_steps_per_epoch=None, save_dir=tempfile.mkdtemp()):
# step 1: prepare dataset
train_data, test_data = load_data()
pipeline = fe.Pipeline(
train_data=train_data,
test_data=test_data,
batch_size=batch_size,
ops=[
Normalize(inputs="x", outputs="x", mean=(0.4914, 0.4822, 0.4465), std=(0.2471, 0.2435, 0.2616)),
PadIfNeeded(min_height=40, min_width=40, image_in="x", image_out="x", mode="train"),
RandomCrop(32, 32, image_in="x", image_out="x", mode="train"),
Sometimes(HorizontalFlip(image_in="x", image_out="x", mode="train")),
CoarseDropout(inputs="x", outputs="x", mode="train", max_holes=1),
ChannelTranspose(inputs="x", outputs="x"),
Onehot(inputs="y", outputs="y", mode="train", num_classes=10, label_smoothing=0.2)
])
# step 2: prepare network
model = fe.build(model_fn=FastCifar, optimizer_fn="adam")
network = fe.Network(ops=[
ModelOp(model=model, inputs="x", outputs="y_pred"),
CrossEntropy(inputs=("y_pred", "y"), outputs="ce"),
UpdateOp(model=model, loss_name="ce")
])
# step 3 prepare estimator
traces = [
Accuracy(true_key="y", pred_key="y_pred"),
BestModelSaver(model=model, save_dir=save_dir, metric="accuracy", save_best_mode="max"),
LRScheduler(model=model, lr_fn=lr_schedule)
]
estimator = fe.Estimator(pipeline=pipeline,
network=network,
epochs=epochs,
traces=traces,
max_train_steps_per_epoch=max_train_steps_per_epoch)
return estimator
def get_estimator(weight=10.0,
epochs=200,
batch_size=1,
max_train_steps_per_epoch=None,
save_dir=tempfile.mkdtemp(),
data_dir=None):
train_data, _ = load_data(batch_size=batch_size, root_dir=data_dir)
device = "cuda" if torch.cuda.is_available() else "cpu"
pipeline = fe.Pipeline(train_data=train_data,
ops=[
ReadImage(inputs=["A", "B"], outputs=["A",
"B"]),
Normalize(inputs=["A", "B"],
outputs=["real_A", "real_B"],
mean=1.0,
std=1.0,
max_pixel_value=127.5),
Resize(height=286,
width=286,
image_in="real_A",
image_out="real_A",
mode="train"),
RandomCrop(height=256,
width=256,
image_in="real_A",
image_out="real_A",
mode="train"),
Resize(height=286,
width=286,
image_in="real_B",
image_out="real_B",
mode="train"),
RandomCrop(height=256,
width=256,
image_in="real_B",
image_out="real_B",
mode="train"),
Sometimes(
HorizontalFlip(image_in="real_A",
image_out="real_A",
mode="train")),
Sometimes(
HorizontalFlip(image_in="real_B",
image_out="real_B",
mode="train")),
ChannelTranspose(inputs=["real_A", "real_B"],
outputs=["real_A", "real_B"])
])
g_AtoB = fe.build(model_fn=Generator,
model_name="g_AtoB",
optimizer_fn=lambda x: torch.optim.Adam(
x, lr=2e-4, betas=(0.5, 0.999)))
g_BtoA = fe.build(model_fn=Generator,
model_name="g_BtoA",
optimizer_fn=lambda x: torch.optim.Adam(
x, lr=2e-4, betas=(0.5, 0.999)))
d_A = fe.build(model_fn=Discriminator,
model_name="d_A",
optimizer_fn=lambda x: torch.optim.Adam(
x, lr=2e-4, betas=(0.5, 0.999)))
d_B = fe.build(model_fn=Discriminator,
model_name="d_B",
optimizer_fn=lambda x: torch.optim.Adam(
x, lr=2e-4, betas=(0.5, 0.999)))
network = fe.Network(ops=[
ModelOp(inputs="real_A", model=g_AtoB, outputs="fake_B"),
ModelOp(inputs="real_B", model=g_BtoA, outputs="fake_A"),
Buffer(image_in="fake_A", image_out="buffer_fake_A"),
Buffer(image_in="fake_B", image_out="buffer_fake_B"),
ModelOp(inputs="real_A", model=d_A, outputs="d_real_A"),
ModelOp(inputs="fake_A", model=d_A, outputs="d_fake_A"),
ModelOp(inputs="buffer_fake_A", model=d_A, outputs="buffer_d_fake_A"),
ModelOp(inputs="real_B", model=d_B, outputs="d_real_B"),
ModelOp(inputs="fake_B", model=d_B, outputs="d_fake_B"),
ModelOp(inputs="buffer_fake_B", model=d_B, outputs="buffer_d_fake_B"),
ModelOp(inputs="real_A", model=g_BtoA, outputs="same_A"),
ModelOp(inputs="fake_B", model=g_BtoA, outputs="cycled_A"),
ModelOp(inputs="real_B", model=g_AtoB, outputs="same_B"),
ModelOp(inputs="fake_A", model=g_AtoB, outputs="cycled_B"),
GLoss(inputs=("real_A", "d_fake_B", "cycled_A", "same_A"),
weight=weight,
device=device,
outputs="g_AtoB_loss"),
GLoss(inputs=("real_B", "d_fake_A", "cycled_B", "same_B"),
weight=weight,
device=device,
outputs="g_BtoA_loss"),
DLoss(inputs=("d_real_A", "buffer_d_fake_A"),
outputs="d_A_loss",
device=device),
DLoss(inputs=("d_real_B", "buffer_d_fake_B"),
outputs="d_B_loss",
device=device),
UpdateOp(model=g_AtoB, loss_name="g_AtoB_loss"),
UpdateOp(model=g_BtoA, loss_name="g_BtoA_loss"),
UpdateOp(model=d_A, loss_name="d_A_loss"),
UpdateOp(model=d_B, loss_name="d_B_loss")
])
traces = [
ModelSaver(model=g_AtoB, save_dir=save_dir, frequency=10),
ModelSaver(model=g_BtoA, save_dir=save_dir, frequency=10),
LRScheduler(model=g_AtoB, lr_fn=lr_schedule),
LRScheduler(model=g_BtoA, lr_fn=lr_schedule),
LRScheduler(model=d_A, lr_fn=lr_schedule),
LRScheduler(model=d_B, lr_fn=lr_schedule)
]
estimator = fe.Estimator(
network=network,
pipeline=pipeline,
epochs=epochs,
traces=traces,
max_train_steps_per_epoch=max_train_steps_per_epoch)
return estimator
def get_estimator(epochs=200,
batch_size=128,
max_train_steps_per_epoch=None,
max_eval_steps_per_epoch=None,
save_dir=tempfile.mkdtemp(),
data_dir=None):
# step 1. prepare pipeline
train_data, eval_data = omniglot.load_data(root_dir=data_dir)
test_data = eval_data.split(0.5)
pipeline = fe.Pipeline(train_data=train_data,
eval_data=eval_data,
test_data=test_data,
batch_size=batch_size,
ops=[
ReadImage(inputs="x_a",
outputs="x_a",
color_flag='gray'),
ReadImage(inputs="x_b",
outputs="x_b",
color_flag='gray'),
Sometimes(ShiftScaleRotate(image_in="x_a",
image_out="x_a",
shift_limit=0.05,
scale_limit=0.2,
rotate_limit=10,
mode="train"),
prob=0.89),
Sometimes(ShiftScaleRotate(image_in="x_b",
image_out="x_b",
shift_limit=0.05,
scale_limit=0.2,
rotate_limit=10,
mode="train"),
prob=0.89),
Minmax(inputs="x_a", outputs="x_a"),
Minmax(inputs="x_b", outputs="x_b")
])
# step 2. prepare model
model = fe.build(model_fn=siamese_network,
model_name="siamese_net",
optimizer_fn="adam")
network = fe.Network(ops=[
ModelOp(inputs=["x_a", "x_b"], model=model, outputs="y_pred"),
CrossEntropy(inputs=("y_pred", "y"), outputs="loss", form="binary"),
UpdateOp(model=model, loss_name="loss")
])
# step 3.prepare estimator
traces = [
LRScheduler(model=model, lr_fn=lr_schedule),
Accuracy(true_key="y", pred_key="y_pred"),
OneShotAccuracy(dataset=eval_data,
model=model,
output_name='one_shot_accuracy'),
BestModelSaver(model=model,
save_dir=save_dir,
metric="one_shot_accuracy",
save_best_mode="max"),
EarlyStopping(monitor="one_shot_accuracy",
patience=20,
compare='max',
mode="eval")
]
estimator = fe.Estimator(
network=network,
pipeline=pipeline,
epochs=epochs,
traces=traces,
max_train_steps_per_epoch=max_train_steps_per_epoch,
max_eval_steps_per_epoch=max_eval_steps_per_epoch)
return estimator
def get_estimator(data_dir=None,
epochs=12,
batch_size_per_gpu=4,
im_size=1344,
model_dir=tempfile.mkdtemp(),
train_steps_per_epoch=None,
eval_steps_per_epoch=None):
assert im_size % 32 == 0, "im_size must be a multiple of 32"
num_device = get_num_devices()
train_ds, val_ds = mscoco.load_data(root_dir=data_dir, load_masks=True)
batch_size = num_device * batch_size_per_gpu
pipeline = fe.Pipeline(
train_data=train_ds,
eval_data=val_ds,
test_data=val_ds,
ops=[
ReadImage(inputs="image", outputs="image"),
MergeMask(inputs="mask", outputs="mask"),
GetImageSize(inputs="image", outputs="imsize", mode="test"),
LongestMaxSize(max_size=im_size,
image_in="image",
mask_in="mask",
bbox_in="bbox",
bbox_params="coco"),
RemoveIf(fn=lambda x: len(x) == 0, inputs="bbox"),
PadIfNeeded(min_height=im_size,
min_width=im_size,
image_in="image",
mask_in="mask",
bbox_in="bbox",
bbox_params="coco",
border_mode=cv2.BORDER_CONSTANT,
value=0),
Sometimes(
HorizontalFlip(image_in="image",
mask_in="mask",
bbox_in="bbox",
bbox_params="coco",
mode="train")),
Resize(height=im_size // 4, width=im_size // 4,
image_in='mask'), # downscale mask for memory efficiency
Gt2Target(inputs=("mask", "bbox"),
outputs=("gt_match", "mask", "classes")),
Delete(keys="bbox"),
Delete(keys="image_id", mode="!test"),
Batch(batch_size=batch_size, pad_value=0)
],
num_process=8 * num_device)
init_lr = 1e-2 / 16 * batch_size
model = fe.build(
model_fn=SoloV2,
optimizer_fn=lambda x: torch.optim.SGD(x, lr=init_lr, momentum=0.9))
network = fe.Network(ops=[
Normalize(inputs="image",
outputs="image",
mean=(0.485, 0.456, 0.406),
std=(0.229, 0.224, 0.225)),
Permute(inputs="image", outputs='image'),
ModelOp(model=model,
inputs="image",
outputs=("feat_seg", "feat_cls_list", "feat_kernel_list")),
LambdaOp(fn=lambda x: x,
inputs="feat_cls_list",
outputs=("cls1", "cls2", "cls3", "cls4", "cls5")),
LambdaOp(fn=lambda x: x,
inputs="feat_kernel_list",
outputs=("k1", "k2", "k3", "k4", "k5")),
Solov2Loss(0,
40,
inputs=("mask", "classes", "gt_match", "feat_seg", "cls1",
"k1"),
outputs=("l_c1", "l_s1")),
Solov2Loss(1,
36,
inputs=("mask", "classes", "gt_match", "feat_seg", "cls2",
"k2"),
outputs=("l_c2", "l_s2")),
Solov2Loss(2,
24,
inputs=("mask", "classes", "gt_match", "feat_seg", "cls3",
"k3"),
outputs=("l_c3", "l_s3")),
Solov2Loss(3,
16,
inputs=("mask", "classes", "gt_match", "feat_seg", "cls4",
"k4"),
outputs=("l_c4", "l_s4")),
Solov2Loss(4,
12,
inputs=("mask", "classes", "gt_match", "feat_seg", "cls5",
"k5"),
outputs=("l_c5", "l_s5")),
CombineLoss(inputs=("l_c1", "l_s1", "l_c2", "l_s2", "l_c3", "l_s3",
"l_c4", "l_s4", "l_c5", "l_s5"),
outputs=("total_loss", "cls_loss", "seg_loss")),
L2Regularizaton(inputs="total_loss",
outputs="total_loss_l2",
model=model,
beta=1e-5,
mode="train"),
UpdateOp(model=model, loss_name="total_loss_l2"),
PointsNMS(inputs="feat_cls_list", outputs="feat_cls_list",
mode="test"),
Predict(inputs=("feat_seg", "feat_cls_list", "feat_kernel_list"),
outputs=("seg_preds", "cate_scores", "cate_labels"),
mode="test")
])
train_steps_epoch = int(np.ceil(len(train_ds) / batch_size))
lr_schedule = {
1:
LRScheduler(
model=model,
lr_fn=lambda step: lr_schedule_warmup(step, init_lr=init_lr)),
2:
LRScheduler(
model=model,
lr_fn=lambda step: cosine_decay(step,
cycle_length=train_steps_epoch *
(epochs - 1),
init_lr=init_lr,
min_lr=init_lr / 100,
start=train_steps_epoch))
}
traces = [
EpochScheduler(lr_schedule),
COCOMaskmAP(data_dir=val_ds.root_dir,
inputs=("seg_preds", "cate_scores", "cate_labels",
"image_id", "imsize"),
mode="test"),
BestModelSaver(model=model, save_dir=model_dir, metric="total_loss")
]
estimator = fe.Estimator(pipeline=pipeline,
network=network,
epochs=epochs,
traces=traces,
monitor_names=("cls_loss", "seg_loss"),
train_steps_per_epoch=train_steps_per_epoch,
eval_steps_per_epoch=eval_steps_per_epoch)
return estimator
def get_estimator(weight=10.0,
epochs=200,
batch_size=1,
train_steps_per_epoch=None,
save_dir=tempfile.mkdtemp(),
data_dir=None):
train_data, _ = load_data(batch_size=batch_size, root_dir=data_dir)
pipeline = fe.Pipeline(
train_data=train_data,
ops=[
ReadImage(inputs=["A", "B"], outputs=["A", "B"]),
Normalize(inputs=["A", "B"], outputs=["real_A", "real_B"], mean=1.0, std=1.0, max_pixel_value=127.5),
Resize(height=286, width=286, image_in="real_A", image_out="real_A", mode="train"),
RandomCrop(height=256, width=256, image_in="real_A", image_out="real_A", mode="train"),
Resize(height=286, width=286, image_in="real_B", image_out="real_B", mode="train"),
RandomCrop(height=256, width=256, image_in="real_B", image_out="real_B", mode="train"),
Sometimes(HorizontalFlip(image_in="real_A", image_out="real_A", mode="train")),
Sometimes(HorizontalFlip(image_in="real_B", image_out="real_B", mode="train")),
PlaceholderOp(outputs=("index_A", "buffer_A")),
PlaceholderOp(outputs=("index_B", "buffer_B"))
])
g_AtoB = fe.build(model_fn=build_generator, model_name="g_AtoB", optimizer_fn=lambda: tf.optimizers.Adam(2e-4, 0.5))
g_BtoA = fe.build(model_fn=build_generator, model_name="g_BtoA", optimizer_fn=lambda: tf.optimizers.Adam(2e-4, 0.5))
d_A = fe.build(model_fn=build_discriminator, model_name="d_A", optimizer_fn=lambda: tf.optimizers.Adam(2e-4, 0.5))
d_B = fe.build(model_fn=build_discriminator, model_name="d_B", optimizer_fn=lambda: tf.optimizers.Adam(2e-4, 0.5))
network = fe.Network(ops=[
ModelOp(inputs="real_A", model=g_AtoB, outputs="fake_B"),
ModelOp(inputs="real_B", model=g_BtoA, outputs="fake_A"),
Buffer(image_in="fake_A", buffer_in="buffer_A", index_in="index_A", image_out="buffer_fake_A"),
Buffer(image_in="fake_B", buffer_in="buffer_B", index_in="index_B", image_out="buffer_fake_B"),
ModelOp(inputs="real_A", model=d_A, outputs="d_real_A"),
ModelOp(inputs="fake_A", model=d_A, outputs="d_fake_A"),
ModelOp(inputs="buffer_fake_A", model=d_A, outputs="buffer_d_fake_A"),
ModelOp(inputs="real_B", model=d_B, outputs="d_real_B"),
ModelOp(inputs="fake_B", model=d_B, outputs="d_fake_B"),
ModelOp(inputs="buffer_fake_B", model=d_B, outputs="buffer_d_fake_B"),
ModelOp(inputs="real_A", model=g_BtoA, outputs="same_A"),
ModelOp(inputs="fake_B", model=g_BtoA, outputs="cycled_A"),
ModelOp(inputs="real_B", model=g_AtoB, outputs="same_B"),
ModelOp(inputs="fake_A", model=g_AtoB, outputs="cycled_B"),
GLoss(inputs=("real_A", "d_fake_B", "cycled_A", "same_A"), weight=weight, outputs="g_AtoB_loss"),
GLoss(inputs=("real_B", "d_fake_A", "cycled_B", "same_B"), weight=weight, outputs="g_BtoA_loss"),
DLoss(inputs=("d_real_A", "buffer_d_fake_A"), outputs="d_A_loss"),
DLoss(inputs=("d_real_B", "buffer_d_fake_B"), outputs="d_B_loss"),
UpdateOp(model=g_AtoB, loss_name="g_AtoB_loss"),
UpdateOp(model=g_BtoA, loss_name="g_BtoA_loss"),
UpdateOp(model=d_A, loss_name="d_A_loss"),
UpdateOp(model=d_B, loss_name="d_B_loss")
])
traces = [
BufferUpdate(input_name="fake_A",
buffer_size=50,
batch_size=batch_size,
mode="train",
output_name=["buffer_A", "index_A"]),
BufferUpdate(input_name="fake_B",
buffer_size=50,
batch_size=batch_size,
mode="train",
output_name=["buffer_B", "index_B"]),
ModelSaver(model=g_AtoB, save_dir=save_dir, frequency=5),
ModelSaver(model=g_BtoA, save_dir=save_dir, frequency=5),
LRScheduler(model=g_AtoB, lr_fn=lr_schedule),
LRScheduler(model=g_BtoA, lr_fn=lr_schedule),
LRScheduler(model=d_A, lr_fn=lr_schedule),
LRScheduler(model=d_B, lr_fn=lr_schedule)
]
estimator = fe.Estimator(network=network,
pipeline=pipeline,
epochs=epochs,
traces=traces,
train_steps_per_epoch=train_steps_per_epoch)
return estimator
def get_estimator(batch_size=8,
epochs=50,
train_steps_per_epoch=None,
eval_steps_per_epoch=None,
save_dir=tempfile.mkdtemp(),
data_dir=None):
# load CUB200 dataset.
train_data = cub200.load_data(root_dir=data_dir)
eval_data = train_data.split(0.3)
test_data = eval_data.split(0.5)
# step 1, pipeline
pipeline = fe.Pipeline(batch_size=batch_size,
train_data=train_data,
eval_data=eval_data,
test_data=test_data,
ops=[
ReadImage(inputs="image",
outputs="image",
parent_path=train_data.parent_path),
Normalize(inputs="image",
outputs="image",
mean=1.0,
std=1.0,
max_pixel_value=127.5),
ReadMat(file='annotation',
keys="seg",
parent_path=train_data.parent_path),
LongestMaxSize(max_size=512,
image_in="image",
image_out="image",
mask_in="seg",
mask_out="seg"),
PadIfNeeded(min_height=512,
min_width=512,
image_in="image",
image_out="image",
mask_in="seg",
mask_out="seg",
border_mode=cv2.BORDER_CONSTANT,
value=0,
mask_value=0),
ShiftScaleRotate(
image_in="image",
mask_in="seg",
image_out="image",
mask_out="seg",
mode="train",
shift_limit=0.2,
rotate_limit=15.0,
scale_limit=0.2,
border_mode=cv2.BORDER_CONSTANT,
value=0,
mask_value=0),
Sometimes(
HorizontalFlip(image_in="image",
mask_in="seg",
image_out="image",
mask_out="seg",
mode="train")),
Reshape(shape=(512, 512, 1),
inputs="seg",
outputs="seg")
])
# step 2, network
resunet50 = fe.build(model_fn=ResUnet50,
model_name="resunet50",
optimizer_fn=lambda: tf.optimizers.Adam(1e-4))
uncertainty = fe.build(model_fn=UncertaintyLossNet,
model_name="uncertainty",
optimizer_fn=lambda: tf.optimizers.Adam(2e-5))
network = fe.Network(ops=[
ModelOp(inputs='image',
model=resunet50,
outputs=["label_pred", "mask_pred"]),
CrossEntropy(inputs=["label_pred", "label"],
outputs="cls_loss",
form="sparse",
average_loss=False),
CrossEntropy(inputs=["mask_pred", "seg"],
outputs="seg_loss",
form="binary",
average_loss=False),
ModelOp(inputs=["cls_loss", "seg_loss"],
model=uncertainty,
outputs="total_loss"),
ReduceLoss(inputs="total_loss", outputs="total_loss"),
UpdateOp(model=resunet50, loss_name="total_loss"),
UpdateOp(model=uncertainty, loss_name="total_loss")
])
# step 3, estimator
traces = [
Accuracy(true_key="label", pred_key="label_pred"),
Dice(true_key="seg", pred_key='mask_pred'),
BestModelSaver(model=resunet50,
save_dir=save_dir,
metric="total_loss",
save_best_mode="min"),
LRScheduler(model=resunet50,
lr_fn=lambda step: cosine_decay(
step, cycle_length=26400, init_lr=1e-4))
]
estimator = fe.Estimator(network=network,
pipeline=pipeline,
traces=traces,
epochs=epochs,
train_steps_per_epoch=train_steps_per_epoch,
eval_steps_per_epoch=eval_steps_per_epoch,
log_steps=500)
return estimator
def get_estimator(epochs=150,
batch_size=32,
save_dir=tempfile.mkdtemp(),
train_steps_per_epoch=None,
eval_steps_per_epoch=None):
# step 1: prepare dataset
train_data, eval_data = load_data()
pipeline = fe.Pipeline(train_data=train_data,
eval_data=eval_data,
batch_size=batch_size * get_num_devices(),
ops=[
Normalize(inputs="x",
outputs="x",
mean=(0.4914, 0.4822, 0.4465),
std=(0.2471, 0.2435, 0.2616)),
PadIfNeeded(min_height=40,
min_width=40,
image_in="x",
image_out="x",
mode="train"),
RandomCrop(32,
32,
image_in="x",
image_out="x",
mode="train"),
Sometimes(
HorizontalFlip(image_in="x",
image_out="x",
mode="train")),
CoarseDropout(inputs="x",
outputs="x",
mode="train",
max_holes=1)
])
# step 2: prepare network
model = fe.build(
model_fn=lambda: pyramidnet_cifar(inputs_shape=(32, 32, 3),
depth=272,
alpha=200,
num_classes=10,
bottleneck=True),
optimizer_fn=lambda: tfa.optimizers.SGDW(
weight_decay=0.0001, lr=0.1, momentum=0.9))
network = fe.Network(ops=[
ModelOp(model=model, inputs="x", outputs="y_pred"),
CrossEntropy(inputs=("y_pred", "y"), outputs="ce", from_logits=True),
UpdateOp(model=model, loss_name="ce")
])
# step 3 prepare estimator
traces = [
Accuracy(true_key="y", pred_key="y_pred"),
LRScheduler(model=model, lr_fn=lr_schedule),
BestModelSaver(model=model,
save_dir=save_dir,
metric="accuracy",
save_best_mode="max")
]
estimator = fe.Estimator(pipeline=pipeline,
network=network,
epochs=epochs,
traces=traces,
train_steps_per_epoch=train_steps_per_epoch,
eval_steps_per_epoch=eval_steps_per_epoch)
return estimator
def get_estimator(data_dir=None,
model_dir=tempfile.mkdtemp(),
epochs=200,
batch_size_per_gpu=32,
train_steps_per_epoch=None,
eval_steps_per_epoch=None):
num_device = get_num_devices()
train_ds, val_ds = mscoco.load_data(root_dir=data_dir)
train_ds = PreMosaicDataset(mscoco_ds=train_ds)
batch_size = num_device * batch_size_per_gpu
pipeline = fe.Pipeline(
train_data=train_ds,
eval_data=val_ds,
ops=[
ReadImage(inputs=("image1", "image2", "image3", "image4"),
outputs=("image1", "image2", "image3", "image4"),
mode="train"),
ReadImage(inputs="image", outputs="image", mode="eval"),
LongestMaxSize(max_size=640,
image_in="image1",
bbox_in="bbox1",
bbox_params=BboxParams("coco", min_area=1.0),
mode="train"),
LongestMaxSize(max_size=640,
image_in="image2",
bbox_in="bbox2",
bbox_params=BboxParams("coco", min_area=1.0),
mode="train"),
LongestMaxSize(max_size=640,
image_in="image3",
bbox_in="bbox3",
bbox_params=BboxParams("coco", min_area=1.0),
mode="train"),
LongestMaxSize(max_size=640,
image_in="image4",
bbox_in="bbox4",
bbox_params=BboxParams("coco", min_area=1.0),
mode="train"),
LongestMaxSize(max_size=640,
image_in="image",
bbox_in="bbox",
bbox_params=BboxParams("coco", min_area=1.0),
mode="eval"),
PadIfNeeded(min_height=640,
min_width=640,
image_in="image",
bbox_in="bbox",
bbox_params=BboxParams("coco", min_area=1.0),
mode="eval",
border_mode=cv2.BORDER_CONSTANT,
value=(114, 114, 114)),
CombineMosaic(inputs=("image1", "image2", "image3", "image4",
"bbox1", "bbox2", "bbox3", "bbox4"),
outputs=("image", "bbox"),
mode="train"),
CenterCrop(height=640,
width=640,
image_in="image",
bbox_in="bbox",
bbox_params=BboxParams("coco", min_area=1.0),
mode="train"),
Sometimes(
HorizontalFlip(image_in="image",
bbox_in="bbox",
bbox_params=BboxParams("coco", min_area=1.0),
mode="train")),
HSVAugment(inputs="image", outputs="image", mode="train"),
ToArray(inputs="bbox", outputs="bbox", dtype="float32"),
CategoryID2ClassID(inputs="bbox", outputs="bbox"),
GTBox(inputs="bbox",
outputs=("gt_sbbox", "gt_mbbox", "gt_lbbox"),
image_size=640),
Delete(keys=("image1", "image2", "image3", "image4", "bbox1",
"bbox2", "bbox3", "bbox4", "bbox"),
mode="train"),
Delete(keys="image_id", mode="eval"),
Batch(batch_size=batch_size, pad_value=0)
])
init_lr = 1e-2 / 64 * batch_size
model = fe.build(
lambda: YoloV5(w=640, h=640, c=3),
optimizer_fn=lambda x: torch.optim.SGD(
x, lr=init_lr, momentum=0.937, weight_decay=0.0005, nesterov=True),
mixed_precision=True)
network = fe.Network(ops=[
RescaleTranspose(inputs="image", outputs="image"),
ModelOp(model=model,
inputs="image",
outputs=("pred_s", "pred_m", "pred_l")),
DecodePred(inputs=("pred_s", "pred_m", "pred_l"),
outputs=("pred_s", "pred_m", "pred_l")),
ComputeLoss(inputs=("pred_s", "gt_sbbox"),
outputs=("sbbox_loss", "sconf_loss", "scls_loss")),
ComputeLoss(inputs=("pred_m", "gt_mbbox"),
outputs=("mbbox_loss", "mconf_loss", "mcls_loss")),
ComputeLoss(inputs=("pred_l", "gt_lbbox"),
outputs=("lbbox_loss", "lconf_loss", "lcls_loss")),
Average(inputs=("sbbox_loss", "mbbox_loss", "lbbox_loss"),
outputs="bbox_loss"),
Average(inputs=("sconf_loss", "mconf_loss", "lconf_loss"),
outputs="conf_loss"),
Average(inputs=("scls_loss", "mcls_loss", "lcls_loss"),
outputs="cls_loss"),
Average(inputs=("bbox_loss", "conf_loss", "cls_loss"),
outputs="total_loss"),
PredictBox(width=640,
height=640,
inputs=("pred_s", "pred_m", "pred_l"),
outputs="box_pred",
mode="eval"),
UpdateOp(model=model, loss_name="total_loss")
])
traces = [
MeanAveragePrecision(num_classes=80,
true_key='bbox',
pred_key='box_pred',
mode="eval"),
BestModelSaver(model=model,
save_dir=model_dir,
metric='mAP',
save_best_mode="max")
]
lr_schedule = {
1:
LRScheduler(model=model,
lr_fn=lambda step: lr_schedule_warmup(
step,
train_steps_epoch=np.ceil(len(train_ds) / batch_size),
init_lr=init_lr)),
4:
LRScheduler(model=model,
lr_fn=lambda epoch: cosine_decay(epoch,
cycle_length=epochs - 3,
init_lr=init_lr,
min_lr=init_lr / 100,
start=4))
}
traces.append(EpochScheduler(lr_schedule))
estimator = fe.Estimator(
pipeline=pipeline,
network=network,
epochs=epochs,
traces=traces,
monitor_names=["bbox_loss", "conf_loss", "cls_loss"],
train_steps_per_epoch=train_steps_per_epoch,
eval_steps_per_epoch=eval_steps_per_epoch)
return estimator
def get_estimator(data_dir=None,
model_dir=tempfile.mkdtemp(),
batch_size=16,
epochs=13,
max_train_steps_per_epoch=None,
max_eval_steps_per_epoch=None,
image_size=512,
num_classes=90):
# pipeline
train_ds, eval_ds = mscoco.load_data(root_dir=data_dir)
pipeline = fe.Pipeline(
train_data=train_ds,
eval_data=eval_ds,
batch_size=batch_size,
ops=[
ReadImage(inputs="image", outputs="image"),
LongestMaxSize(image_size,
image_in="image",
image_out="image",
bbox_in="bbox",
bbox_out="bbox",
bbox_params=BboxParams("coco", min_area=1.0)),
PadIfNeeded(
image_size,
image_size,
border_mode=cv2.BORDER_CONSTANT,
image_in="image",
image_out="image",
bbox_in="bbox",
bbox_out="bbox",
bbox_params=BboxParams("coco", min_area=1.0),
),
Sometimes(
HorizontalFlip(mode="train",
image_in="image",
image_out="image",
bbox_in="bbox",
bbox_out="bbox",
bbox_params='coco')),
# normalize from uint8 to [-1, 1]
Normalize(inputs="image",
outputs="image",
mean=1.0,
std=1.0,
max_pixel_value=127.5),
ShiftLabel(inputs="bbox", outputs="bbox"),
AnchorBox(inputs="bbox",
outputs="anchorbox",
width=image_size,
height=image_size),
ChannelTranspose(inputs="image", outputs="image")
],
pad_value=0)
# network
model = fe.build(model_fn=lambda: RetinaNet(num_classes=num_classes),
optimizer_fn=lambda x: torch.optim.SGD(
x, lr=2e-4, momentum=0.9, weight_decay=0.0001))
network = fe.Network(ops=[
ModelOp(model=model, inputs="image", outputs=["cls_pred", "loc_pred"]),
RetinaLoss(inputs=["anchorbox", "cls_pred", "loc_pred"],
outputs=["total_loss", "focal_loss", "l1_loss"]),
UpdateOp(model=model, loss_name="total_loss"),
PredictBox(input_shape=(image_size, image_size, 3),
inputs=["cls_pred", "loc_pred"],
outputs="pred",
mode="eval")
])
# estimator
traces = [
LRScheduler(model=model, lr_fn=lr_fn),
BestModelSaver(model=model,
save_dir=model_dir,
metric='mAP',
save_best_mode="max"),
MeanAveragePrecision(num_classes=num_classes,
true_key='bbox',
pred_key='pred',
mode="eval")
]
estimator = fe.Estimator(
pipeline=pipeline,
network=network,
epochs=epochs,
traces=traces,
max_train_steps_per_epoch=max_train_steps_per_epoch,
max_eval_steps_per_epoch=max_eval_steps_per_epoch,
monitor_names=["l1_loss", "focal_loss"])
return estimator
