/ Machine Learning

Shape inference for PyTorch (like in Keras) + new SoTA layers

Shape inference for PyTorch (like in Keras) + new SoTA layers

torchlayers

torchlayers is a library based on PyTorch providing automatic shape and dimensionality inference of torch.nn layers + additional building blocks featured in current SOTA architectures (e.g. Efficient-Net).

Above requires no user intervention (except single call to torchlayers.build) similarly to the one seen in Keras.

Main functionalities:

  • Shape inference for most of torch.nn module (convolutional, recurrent, transformer, attention and linear layers)
  • Dimensionality inference (e.g. torchlayers.Conv working as torch.nn.Conv1d/2d/3d based on input shape)
  • Shape inference of custom modules (see examples section)
  • Additional Keras-like layers (e.g. torchlayers.Reshape or torchlayers.StandardNormalNoise)
  • Additional SOTA layers mostly from ImageNet competitions
    (e.g. PolyNet,
    Squeeze-And-Excitation,
    StochasticDepth)
  • Useful defaults ("same" padding and default kernel_size=3 for Conv, dropout rates etc.)
  • Zero overhead and torchscript support

Examples

For full functionality please check torchlayers documentation.
Below examples should introduce all necessary concepts you should know.

Simple convolutional image and text classifier

  • We will use single "model" for both tasks.
    Firstly let's define it using torch.nn and torchlayers:
import torch
import torchlayers

# torch.nn and torchlayers can be mixed easily
model = torch.nn.Sequential(
    torchlayers.Conv(64),  # specify ONLY out_channels
    torch.nn.ReLU(),  # use torch.nn wherever you wish
    torchlayers.BatchNorm(),  # BatchNormNd inferred from input
    torchlayers.Conv(128),  # Default kernel_size equal to 3
    torchlayers.ReLU(),
    torchlayers.Conv(256, kernel_size=11),  # "same" padding as default
    torchlayers.GlobalMaxPool(),  # Known from Keras
    torchlayers.Linear(10),  # Output for 10 classes
)

print(model)

Above would give you model's summary like this (notice question marks for not yet inferred values):

Sequential(
  (0): Conv(in_channels=?, out_channels=64, kernel_size=3, stride=1, padding=same, dilation=1, groups=1, bias=True, padding_mode=zeros)
  (1): ReLU()
  (2): BatchNorm(num_features=?, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (3): Conv(in_channels=?, out_channels=128, kernel_size=3, stride=1, padding=same, dilation=1, groups=1, bias=True, padding_mode=zeros)
  (4): ReLU()
  (5): Conv(in_channels=?, out_channels=256, kernel_size=11, stride=1, padding=same, dilation=1, groups=1, bias=True, padding_mode=zeros)
  (6): GlobalMaxPool()
  (7): Linear(in_features=?, out_features=10, bias=True)
)
  • Now you can build/instantiate your model with example input (in this case MNIST-like):
mnist_model = torchlayers.build(model, torch.randn(1, 3, 28, 28))
  • Or if it's text classification you are after, same model could be built with different
    input shape (e.g. for text classification using 300 dimensional pretrained embedding):
# [batch, embedding, timesteps], first dimension > 1 for BatchNorm1d to work
text_model = torchlayers.build(model, torch.randn(2, 300, 1))
  • Finally, you can print both models after instantiation, provided below side
    by-side for readability (notice different dimenstionality, e.g. Conv2d vs Conv1d after torchlayers.build):
                # MNIST CLASSIFIER                TEXT CLASSIFIER

                Sequential(                       Sequential(
                  (0): Conv1d(300, 64)              (0): Conv2d(3, 64)
                  (1): ReLU()                       (1): ReLU()
                  (2): BatchNorm1d(64)              (2): BatchNorm2d(64)
                  (3): Conv1d(64, 128)              (3): Conv2d(64, 128)
                  (4): ReLU()                       (4): ReLU()
                  (5): Conv1d(128, 256)             (5): Conv2d(128, 256)
                  (6): GlobalMaxPool()              (6): GlobalMaxPool()
                  (7): Linear(256, 10)              (7): Linear(256, 10)
                )                                 )

As you can see both modules "compiled" into original pytorch layers.

Custom modules with shape inference capabilities

User can define any module and make it shape inferable with torchlayers.infer
function:

 # Class defined with in_features
 # It might be a good practice to use _ prefix and Impl as postfix
 # to differentiate from shape inferable version
class _MyLinearImpl(torch.nn.Module):
    def __init__(self, in_features: int, out_features: int):
        super().__init__()
        self.weight = torch.nn.Parameter(torch.randn(out_features, in_features))
        self.bias = torch.nn.Parameter(torch.randn(out_features))

    def forward(self, inputs):
        return torch.nn.functional.linear(inputs, self.weight, self.bias)

MyLinear = torchlayers.infer(_MyLinearImpl)

# Build and use just like any other layer in this library
layer =torchlayers.build(MyLinear(out_features=32), torch.randn(1, 64))
layer(torch.randn(1, 64))

By default inputs.shape[1] will be used as in_features value
during initial forward pass. If you wish to use different index (e.g. to infer using
inputs.shape[3]) use MyLayer = torchlayers.infer(_MyLayerImpl, index=3) as a decorator.

Autoencoder with inverted residual bottleneck and pixel shuffle

Please check code comments and documentation
if needed. If you are unsure what autoencoder is you could see
this example blog post.

Below is a convolutional denoising autoencoder example for ImageNet-like images.
Think of it like a demonstration of capabilities of different layers
and building blocks provided by torchlayers.

# Input - 3 x 256 x 256 for ImageNet reconstruction
class AutoEncoder(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.encoder = torchlayers.Sequential(
            torchlayers.StandardNormalNoise(),  # Apply noise to input images
            torchlayers.Conv(64, kernel_size=7),
            torchlayers.activations.Swish(),  # Direct access to module .activations
            torchlayers.InvertedResidualBottleneck(squeeze_excitation=False),
            torchlayers.AvgPool(),  # shape 64 x 128 x 128, kernel_size=2 by default
            torchlayers.HardSwish(),  # Access simply through torchlayers
            torchlayers.SeparableConv(128),  # Up number of channels to 128
            torchlayers.InvertedResidualBottleneck(),  # Default with squeeze excitation
            torch.nn.ReLU(),
            torchlayers.AvgPool(),  # shape 128 x 64 x 64, kernel_size=2 by default
            torchlayers.DepthwiseConv(256),  # DepthwiseConv easier to use
            # Pass input thrice through the same weights like in PolyNet
            torchlayers.Poly(torchlayers.InvertedResidualBottleneck(), order=3),
            torchlayers.ReLU(),  # all torch.nn can be accessed via torchlayers
            torchlayers.MaxPool(),  # shape 256 x 32 x 32
            torchlayers.Fire(out_channels=512),  # shape 512 x 32 x 32
            torchlayers.SqueezeExcitation(hidden=64),
            torchlayers.InvertedResidualBottleneck(),
            torchlayers.MaxPool(),  # shape 512 x 16 x 16
            torchlayers.InvertedResidualBottleneck(squeeze_excitation=False),
            # Randomly switch off the last two layers with 0.5 probability
            torchlayers.StochasticDepth(
                torch.nn.Sequential(
                    torchlayers.InvertedResidualBottleneck(squeeze_excitation=False),
                    torchlayers.InvertedResidualBottleneck(squeeze_excitation=False),
                ),
                p=0.5,
            ),
            torchlayers.AvgPool(),  # shape 512 x 8 x 8
        )

        # This one is more "standard"
        self.decoder = torchlayers.Sequential(
            torchlayers.Poly(torchlayers.InvertedResidualBottleneck(), order=2),
            # Has ICNR initialization by default after calling `build`
            torchlayers.ConvPixelShuffle(out_channels=512, upscale_factor=2),
            # Shape 512 x 16 x 16 after PixelShuffle
            torchlayers.Poly(torchlayers.InvertedResidualBottleneck(), order=3),
            torchlayers.ConvPixelShuffle(out_channels=256, upscale_factor=2),
            # Shape 256 x 32 x 32
            torchlayers.Poly(torchlayers.InvertedResidualBottleneck(), order=3),
            torchlayers.ConvPixelShuffle(out_channels=128, upscale_factor=2),
            # Shape 128 x 64 x 64
            torchlayers.Poly(torchlayers.InvertedResidualBottleneck(), order=4),
            torchlayers.ConvPixelShuffle(out_channels=64, upscale_factor=2),
            # Shape 64 x 128 x 128
            torchlayers.InvertedResidualBottleneck(),
            torchlayers.Conv(256),
            torchlayers.Dropout(),  # Defaults to 0.5 and Dropout2d for images
            torchlayers.Swish(),
            torchlayers.InstanceNorm(),
            torchlayers.ConvPixelShuffle(out_channels=32, upscale_factor=2),
            # Shape 32 x 256 x 256
            torchlayers.Conv(16),
            torchlayers.Swish(),
            torchlayers.Conv(3),
            # Shape 3 x 256 x 256
        )

    def forward(self, inputs):
        return self.decoder(self.encoder(inputs))

Now one can instantiate the module and use it with torch.nn.MSELoss as per usual.

autoencoder = torchlayers.build(AutoEncoder(), torch.randn(1, 3, 256, 256))

Installation

pip

Latest release:

pip install --user torchlayers

Nightly:

pip install --user torchlayers-nightly

Docker

CPU standalone and various versions of GPU enabled images are available
at dockerhub.

For CPU quickstart, issue:

docker pull szymonmaszke/torchlayers:18.04

Nightly builds are also available, just prefix tag with nightly_. If you are going for GPU image make sure you have
nvidia/docker installed and it's runtime set.

GitHub

Comments