torch-activation 1.0.0

Activation functions for PyTorch

PyTorch Activations

PyTorch Activations is a collection of activation functions for the PyTorch library. This project aims to provide an easy-to-use solution for experimenting with different activation functions or simply adding variety to your models.

Installation

You can install PyTorch Activations using pip:

$ pip install torch-activation

Usage

To use the activation functions, import them from torch_activation. Here's an example:

import torch_activation as tac

m = tac.ShiLU(inplace=True)
x = torch.rand(16, 3, 384, 384)
m(x)

Or in nn.Sequential:

import torch
import torch.nn as nn
import torch_activation as tac

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.net = nn.Sequential(
            nn.Conv2d(64, 32, 2),
            tac.DELU(),
            nn.ConvTranspose2d(32, 64, 2),
            tac.ReLU(inplace=True),
        )

    def forward(self, x):
        return self.net(x)

Activation functions can be imported directly from the package, such as torch_activation.CoLU, or from submodules, such as torch_activation.classical.CoLU or torch_activation.classical.sigmoid_weighted.CoLU.

To learn more about usage and the comprehended list of available functions, please refer to Documentation

We hope you find PyTorch Activations useful for your experimentation and model development. Enjoy exploring different activation functions!

Contact

Alan Huynh - LinkedIn - [e-mail]

Project Link: https://github.com/hdmquan/torch_activation

Documentation Link: https://torch-activation.readthedocs.io

PyPI Link: https://pypi.org/project/torch-activation/