complexNN 0.1.2

A toolbox for using complex valued standard network modules in PyTorch.

ComplexNN: Complex Neural Network Modules

Since the latest versions of PyTorch support matrix operations and gradient descent on plural parameters, this repository provides the latest plural form of some standard PyTorch network modules. Compared with utilizing two sets of parameters to represent the real and imaginary parts of the network plural parameters respectively, directly utilizing complex numbers as network parameters will halve the number of trainable parameters, which speeds up the training process.

Install

To install complexNN for the first time:

pip install complexNN

To upgrade a previous installation of complexNN to the most recent version:

pip install --upgrade complexNN

Versions

v0.0.1 Provides the plural form of the base standard PyTorch network module.

v0.1.1 Adds support for the Linear Recurrent Unit (LRU).

v0.1.2 Bug fixed. Adds support for BatchNorm2d, and BatchNorm3d.

Modules

The complex form modules include

complexLayer	complexRNNcell	complexActivation	complexFunction
Linear	RNN Cell	Relu	BatchNorm
	GRU Cell	Gelu	LayerNorm
	LSTM Cell	Tanh	Dropout
	LRU Cell [1]	Sigmoid

Note that the native version of torch.nn.Dropout is supported:exclamation::exclamation: Other modules will be considered for updates in the future.

Examples

Multilayer perceptron

import torch.nn as nn
from complexNN.complexActivation import complexTanh
from complexNN.complexLinear import complexLinear


class complexMLP(nn.Module):
    """
    Complex Multilayer Perceptron
    """

    def __init__(self, input_size, hidden_size, output_size, num_layers):
        super(complexMLP, self).__init__()
        self.num_layers = num_layers
        self.input_layer = complexLinear(input_size, hidden_size)
        self.hidden_layers = nn.ModuleList([complexLinear(hidden_size, hidden_size) for _ in range(num_layers - 1)])
        self.output_layer = complexLinear(hidden_size, output_size)
        self.dropout = nn.Dropout(0.5)

    def forward(self, x):
        x = complexTanh(self.input_layer(x))
        x = self.dropout(x)
        for i in range(self.layer_num - 1):
            x = complexTanh(self.hidden_layers[i](x))
            x = self.dropout(x)
        output = self.output_layer(x)
        return output

Recurrent neural network

import torch.nn as nn
from complexNN.complexRNNcell import complexRNNCell


class complexRNN(nn.Module):
    def __init__(self, input_size, hidden_size, num_layers):
        super(complexRNN, self).__init__()
        self.num_layers = num_layers
        self.rnn_layers = nn.ModuleList()
        for _ in range(num_layers):
            self.rnn_layers.append(complexRNNCell(input_size, hidden_size))
            input_size = hidden_size

    def forward(self, x, h_0):
        h_prev = h_0
        for i in range(self.laryer_num):
            h_prev = self.rnn_layers[i](x, h_prev)
        return h_prev

Cite as

@misc{ComplexNN,
      title={ComplexNN: Complex Neural Network Modules},
      author={Xinyuan Liao},
      Url= {https://github.com/XinyuanLiao/ComplexNN}, 
      year={2023}
}

Reference

[1] Orvieto, Antonio, et al. "Resurrecting recurrent neural networks for long sequences." arXiv preprint arXiv:2303.06349 (2023).