def test_samples(fmodel_and_data: ModelAndData, batchsize: int) -> None:
fmodel, _, _ = fmodel_and_data
if hasattr(fmodel, "data_format"):
data_format = fmodel.data_format # type: ignore
x, y = fbn.samples(fmodel, batchsize=batchsize)
assert len(x) == len(y) == batchsize
assert not ep.istensor(x)
assert not ep.istensor(y)
x, y = fbn.samples(fmodel,
batchsize=batchsize,
data_format=data_format)
assert len(x) == len(y) == batchsize
assert not ep.istensor(x)
assert not ep.istensor(y)
with pytest.raises(ValueError):
data_format = {
"channels_first": "channels_last",
"channels_last": "channels_first",
}[data_format]
fbn.samples(fmodel, batchsize=batchsize, data_format=data_format)
else:
x, y = fbn.samples(fmodel,
batchsize=batchsize,
data_format="channels_first")
assert len(x) == len(y) == batchsize
assert not ep.istensor(x)
assert not ep.istensor(y)
with pytest.raises(ValueError):
fbn.samples(fmodel, batchsize=batchsize)
def jax_simple_model(request: Any) -> ModelAndData:
import jax
def model(x: Any) -> Any:
return jax.numpy.mean(x, axis=(1, 2))
bounds = (0, 1)
fmodel = fbn.JAXModel(model, bounds=bounds)
x, _ = fbn.samples(fmodel,
dataset="imagenet",
batchsize=16,
data_format="channels_last")
x = ep.astensor(x)
y = fmodel(x).argmax(axis=-1)
return fmodel, x, y
def tensorflow_resnet50(request: Any) -> ModelAndData:
if request.config.option.skipslow:
pytest.skip()
import tensorflow as tf
model = tf.keras.applications.ResNet50(weights="imagenet")
preprocessing = dict(flip_axis=-1, mean=[104.0, 116.0,
123.0]) # RGB to BGR
fmodel = fbn.TensorFlowModel(model,
bounds=(0, 255),
preprocessing=preprocessing)
x, y = fbn.samples(fmodel, dataset="imagenet", batchsize=16)
x = ep.astensor(x)
y = ep.astensor(y)
return fmodel, x, y
def tensorflow_simple_functional(request: Any) -> ModelAndData:
import tensorflow as tf
channels = 3
h = w = 224
data_format = tf.keras.backend.image_data_format()
shape = (channels, h, w) if data_format == "channels_first" else (h, w,
channels)
x = x_ = tf.keras.Input(shape=shape)
x = tf.keras.layers.GlobalAveragePooling2D()(x)
model = tf.keras.Model(inputs=x_, outputs=x)
bounds = (0, 1)
fmodel = fbn.TensorFlowModel(model, bounds=bounds)
x, _ = fbn.samples(fmodel, dataset="imagenet", batchsize=16)
x = ep.astensor(x)
y = fmodel(x).argmax(axis=-1)
return fmodel, x, y
def tensorflow_simple_sequential(
device: Optional[str] = None,
preprocessing: fbn.types.Preprocessing = None) -> ModelAndData:
import tensorflow as tf
with tf.device(device):
model = tf.keras.Sequential()
model.add(tf.keras.layers.GlobalAveragePooling2D())
bounds = (0, 1)
fmodel = fbn.TensorFlowModel(model,
bounds=bounds,
device=device,
preprocessing=preprocessing)
x, _ = fbn.samples(fmodel, dataset="imagenet", batchsize=16)
x = ep.astensor(x)
y = fmodel(x).argmax(axis=-1)
return fmodel, x, y
def pytorch_resnet18(request: Any) -> ModelAndData:
if request.config.option.skipslow:
pytest.skip()
import torchvision.models as models
model = models.resnet18(pretrained=True).eval()
preprocessing = dict(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225],
axis=-3)
fmodel = fbn.PyTorchModel(model,
bounds=(0, 1),
preprocessing=preprocessing)
x, y = fbn.samples(fmodel, dataset="imagenet", batchsize=16)
x = ep.astensor(x)
y = ep.astensor(y)
return fmodel, x, y
def tensorflow_simple_subclassing(request: Any) -> ModelAndData:
import tensorflow as tf
class Model(tf.keras.Model): # type: ignore
def __init__(self) -> None:
super().__init__()
self.pool = tf.keras.layers.GlobalAveragePooling2D()
def call(self, x: tf.Tensor) -> tf.Tensor: # type: ignore
x = self.pool(x)
return x
model = Model()
bounds = (0, 1)
fmodel = fbn.TensorFlowModel(model, bounds=bounds)
x, _ = fbn.samples(fmodel, dataset="imagenet", batchsize=16)
x = ep.astensor(x)
y = fmodel(x).argmax(axis=-1)
return fmodel, x, y
def pytorch_simple_model(
device: Any = None,
preprocessing: fbn.types.Preprocessing = None) -> ModelAndData:
import torch
class Model(torch.nn.Module):
def forward(self, x: torch.Tensor) -> torch.Tensor: # type: ignore
x = torch.mean(x, 3)
x = torch.mean(x, 2)
return x
model = Model().eval()
bounds = (0, 1)
fmodel = fbn.PyTorchModel(model,
bounds=bounds,
device=device,
preprocessing=preprocessing)
x, _ = fbn.samples(fmodel, dataset="imagenet", batchsize=16)
x = ep.astensor(x)
y = fmodel(x).argmax(axis=-1)
return fmodel, x, y
def foolbox2_simple_model(channel_axis: int) -> ModelAndData:
class Foolbox2DummyModel(foolbox.models.base.Model): # type: ignore
def __init__(self) -> None:
super().__init__(bounds=(0, 1),
channel_axis=channel_axis,
preprocessing=(0, 1))
def forward(self, inputs: Any) -> Any:
if channel_axis == 1:
return inputs.mean(axis=(2, 3))
elif channel_axis == 3:
return inputs.mean(axis=(1, 2))
def num_classes(self) -> int:
return 3
model = Foolbox2DummyModel()
fmodel = fbn.Foolbox2Model(model)
with pytest.warns(UserWarning):
x, _ = fbn.samples(fmodel, dataset="imagenet", batchsize=16)
x = ep.astensor(x)
y = fmodel(x).argmax(axis=-1)
return fmodel, x, y