torchex library provides advanced Neural Network Layers. You can easily use them like using original pytorch.

$ pip install torchex

• Pytorch >= 1.0

• https://torchex.readthedocs.io/en/latest/index.html

import torch
import torchex.nn as exnn

net = exnn.Linear(10)
# You don't need to give the size of input for this module.
# This network is equivalent to `nn.Linear(100, 10)`.

x = troch.randn(10, 100)

y = net(x)

• torchex.nn.Pass
• torchex.nn.Flatten
• torchex.nn.Linear
  • Lazy style
• torchex.nn.Conv1d
  • Lazy style
• torchex.nn.Conv2d
  • Lazy style
• torchex.nn.Conv3d
  • Lazy style
• torchex.nn.Conv2dLocal
• torchex.nn.GlobalAvgPool1d
• torchex.nn.GlobalAvgPool2d
• torchex.nn.GlobalMaxPool1d
• torchex.nn.GlobalMaxPool2d
• torchex.nn.MaxAvgPool2d
• torch.nn.Crop2d
• torch.nn.Crop3d
• torch.nn.MLPConv2d
• torch.nn.UpsampleConvLayer
• torch.nn.CordConv2d
• torch.nn.DFT1d
• torch.nn.DFT2d
• torch.nn.PeriodicPad2d
• torch.nn.PeriodicPad3d
• torch.nn.Highway
• torch.nn.Inception
• torch.nn.InceptionBN
• torch.nn.IndRNNCell
• torch.nn.IndRNNTanhCell
• torch.nn.IndRNNReLuCell
• torch.nn.IndRNN
• torch.nn.GraphLinear
• torch.nn.GraphConv
• torch.nn.SparseMM
• torch.nn.GraphBatchNrom

• torchex.data.transforms.PadRandomSift
• torchex.data.transforms.RandomResize

for visualization

• torchex.attribute.IntegratedGradients