High-Level Training, Data Augmentation, and Utilities for Pytorch

v0.1.3 JUST RELEASED - contains significant improvements, bug fixes, and additional support. Get it from the releases, or pull the master branch.

This package provides a few things:

A high-level module for Keras-like training with callbacks, constraints, and regularizers.
Comprehensive data augmentation, transforms, sampling, and loading
Utility tensor and variable functions so you don't need numpy as often

Have any feature requests? Submit an issue! I'll make it happen. Specifically, any data augmentation, data loading, or sampling functions.

Want to contribute? Check the issues page for those tagged with [contributions welcome].

ModuleTrainer

The ModuleTrainer class provides a high-level training interface which abstracts away the training loop while providing callbacks, constraints, initializers, regularizers, and more.

Example:

from torchsample.modules import ModuleTrainer# Define your model EXACTLY as normalclass Network(nn.Module):    def __init__(self):        super(Network, self).__init__()        self.conv1 = nn.Conv2d(1, 32, kernel_size=3)        self.conv2 = nn.Conv2d(32, 64, kernel_size=3)        self.fc1 = nn.Linear(1600, 128)        self.fc2 = nn.Linear(128, 10)    def forward(self, x):
        x = F.relu(F.max_pool2d(self.conv1(x), 2))
        x = F.relu(F.max_pool2d(self.conv2(x), 2))
        x = x.view(-1, 1600)
        x = F.relu(self.fc1(x))
        x = F.dropout(x, training=self.training)
        x = self.fc2(x)        return F.log_softmax(x)

model = Network()
trainer = ModuleTrainer(model)

trainer.compile(loss='nll_loss',                optimizer='adadelta')

trainer.fit(x_train, y_train, 
            val_data=(x_test, y_test),            num_epoch=20, 
            batch_size=128,            verbose=1)

You also have access to the standard evaluation and prediction functions:

loss = model.evaluate(x_train, y_train)
y_pred = model.predict(x_train)

Torchsample provides a wide range of callbacks, generally mimicking the interface found in Keras:

EarlyStopping
ModelCheckpoint
LearningRateScheduler
ReduceLROnPlateau
CSVLogger

from torchsample.callbacks import EarlyStopping

callbacks = [EarlyStopping(monitor='val_loss', patience=5)]
model.set_callbacks(callbacks)

Torchsample also provides regularizers:

L1Regularizer
L2Regularizer
L1L2Regularizer

and constraints:

UnitNorm
MaxNorm
NonNeg

Both regularizers and constraints can be selectively applied on layers using regular expressions and the module_filter argument. Constraints can be explicit (hard) constraints applied at an arbitrary batch or epoch frequency, or they can be implicit (soft) constraints similar to regularizers where the the constraint deviation is added as a penalty to the total model loss.

from torchsample.constraints import MaxNorm, NonNegfrom torchsample.regularizers import L1Regularizer# hard constraint applied every 5 batcheshard_constraint = MaxNorm(value=2., frequency=5, unit='batch', module_filter='*fc*')# implicit constraint added as a penalty term to model losssoft_constraint = NonNeg(lagrangian=True, scale=1e-3, module_filter='*fc*')
constraints = [hard_constraint, soft_constraint]
model.set_constraints(constraints)

regularizers = [L1Regularizer(scale=1e-4, module_filter='*conv*')]
model.set_regularizers(regularizers)

You can also fit directly on a torch.utils.data.DataLoader and can have a validation set as well :

from torchsample import TensorDatasetfrom torch.utils.data import DataLoader

train_dataset = TensorDataset(x_train, y_train)
train_loader = DataLoader(train_dataset, batch_size=32)

val_dataset = TensorDataset(x_val, y_val)
val_loader = DataLoader(val_dataset, batch_size=32)

trainer.fit_loader(loader, val_loader=val_loader, num_epoch=100)

Utility Functions

Finally, torchsample provides a few utility functions not commonly found:

Tensor Functions

th_iterproduct (mimics itertools.product)
th_gather_nd (N-dimensional version of torch.gather)
th_random_choice (mimics np.random.choice)
th_pearsonr (mimics scipy.stats.pearsonr)
th_corrcoef (mimics np.corrcoef)
th_affine2d and th_affine3d (affine transforms on torch.Tensors)

Variable Functions

F_affine2d and F_affine3d
F_map_coordinates2d and F_map_coordinates3d

Data Augmentation and Datasets

The torchsample package provides a ton of good data augmentation and transformation tools which can be applied during data loading. The package also provides the flexible TensorDataset and FolderDataset classes to handle most dataset needs.

Torch Transforms

These transforms work directly on torch tensors

Compose()
AddChannel()
SwapDims()
RangeNormalize()
StdNormalize()
Slice2D()
RandomCrop()
SpecialCrop()
Pad()
RandomFlip()
ToTensor()

Affine Transforms

The following transforms perform affine (or affine-like) transforms on torch tensors.

Rotate()
Translate()
Shear()
Zoom()

We also provide a class for stringing multiple affine transformations together so that only one interpolation takes place:

Affine()
AffineCompose()

Datasets and Sampling

We provide the following datasets which provide general structure and iterators for sampling from and using transforms on in-memory or out-of-memory data: