def test_graph_basic(self):
from inferno.extensions.containers.graph import Graph
from inferno.extensions.layers.convolutional import ConvELU2D
from inferno.utils.model_utils import ModelTester
# Build graph
model = Graph()
model.add_input_node('input_0')
model.add_node('conv0', ConvELU2D(1, 10, 3), previous='input_0')
model.add_node('conv1', ConvELU2D(10, 1, 3), previous='conv0')
model.add_output_node('output_0', previous='conv1')
ModelTester((1, 1, 100, 100), (1, 1, 100, 100))(model)
def __init__(self, in_channels, out_channels, size=3, kernel_size=3):
super(ResBlock, self).__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.size = size
conv_ops = [ConvELU2D(kernel_size=kernel_size, in_channels=in_channels, out_channels=out_channels)]
for i in range(size-1):
conv_ops.append(ConvELU2D(kernel_size=kernel_size, in_channels=out_channels, out_channels=out_channels))
self.conv_ops = nn.ModuleList(conv_ops)
def __init__(self, previous_in_channels, out_channels, kernel_size, pre_output):
super(Xcoder, self).__init__()
assert out_channels % 2 == 0
self.in_channels = sum(previous_in_channels)
self.out_channels = out_channels
self.conv1 = ConvELU2D(
in_channels=self.in_channels, out_channels=self.out_channels // 2, kernel_size=kernel_size
)
self.conv2 = ConvELU2D(
in_channels=self.in_channels + (self.out_channels // 2),
out_channels=self.out_channels // 2,
kernel_size=kernel_size,
)
self.pre_output = pre_output
def test_graph_device_transfers(self):
from inferno.extensions.containers.graph import Graph
from inferno.extensions.layers.convolutional import ConvELU2D
import torch
# Build graph
model = Graph()
model.add_input_node('input_0')
model.add_node('conv0', ConvELU2D(1, 10, 3), previous='input_0')
model.add_node('conv1', ConvELU2D(10, 1, 3), previous='conv0')
model.add_output_node('output_0', previous='conv1')
# Transfer
model.to_device('conv0', 'cpu').to_device('conv1', 'cuda', 0)
x = torch.rand(1, 1, 100, 100)
y = model(x)
self.assertIsInstance(y.data, torch.cuda.FloatTensor)
def __init__(self, in_channels, out_channels,out_per_conv=4, size=4, kernel_size=3):
super(DenseBlock, self).__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.size = size
sum_of_channels = int(in_channels)
conv_ops = []
for i in range(size):
if i + 1 == size:
conv_ops.append(ConvELU2D(kernel_size=kernel_size, in_channels=sum_of_channels, out_channels=out_channels))
sum_of_channels += out_per_conv
else:
conv_ops.append(ConvELU2D(kernel_size=kernel_size, in_channels=sum_of_channels, out_channels=out_per_conv))
sum_of_channels += out_per_conv
self.conv_ops = nn.ModuleList(conv_ops)
from inferno.io.box.cifar10 import get_cifar10_loaders
from inferno.trainers.basic import Trainer
from inferno.trainers.callbacks.logging.tensorboard import TensorboardLogger
from inferno.extensions.layers.convolutional import ConvELU2D
from inferno.extensions.layers.reshape import Flatten
# Fill these in:
LOG_DIRECTORY = 'logs'
SAVE_DIRECTORY = 'models'
DATASET_DIRECTORY = 'data'
DOWNLOAD_CIFAR = True
USE_CUDA = False
# Build torch model
model = nn.Sequential(
ConvELU2D(in_channels=3, out_channels=256, kernel_size=3),
nn.MaxPool2d(kernel_size=2, stride=2),
ConvELU2D(in_channels=256, out_channels=256, kernel_size=3),
nn.MaxPool2d(kernel_size=2, stride=2),
ConvELU2D(in_channels=256, out_channels=256, kernel_size=3),
nn.MaxPool2d(kernel_size=2, stride=2), Flatten(),
nn.Linear(in_features=(256 * 4 * 4), out_features=10), nn.Softmax())
# Load loaders
train_loader, validate_loader = get_cifar10_loaders(DATASET_DIRECTORY,
download=DOWNLOAD_CIFAR)
# Build trainer
trainer = Trainer(model) \
.build_criterion('CrossEntropyLoss') \
.build_metric('CategoricalError') \