nninit

Weight initialization schemes for PyTorch nn.Modules. This is a port of the popular nninit for Torch7 by @kaixhin.

Update

This repo has been merged into PyTorch's nn module, I recommend you use that version going forward.

PyTorch Example

import nninitfrom torch import nnimport torch.nn.init as initimport numpy as npclass Net(nn.Module):
  def __init__(self):
     super(Net, self).__init__()
     self.conv1 = nn.Conv2d(5, 10, (3, 3))
     init.xavier_uniform(self.conv1.weight, gain=np.sqrt(2))
     init.constant(self.conv1.bias, 0.1)network = Net()

Installation

Clone the repo and run python setup install

Usage

import nninitfrom torch import nnimport numpy as npclass Net(nn.Module):
  def __init__(self):
     super(Net, self).__init__()
     self.conv1 = nn.Conv2d(5, 10, (3, 3))
     nninit.xavier_uniform(self.conv1.weight, gain=np.sqrt(2))
     nninit.constant(self.conv1.bias, 0.1)network = Net()

Supported Schemes

• nninit.uniform(tensor, a=0, b=1) - Fills tensor with values from a uniform, U(a,b)

• nninit.normal(tensor, mean=0, std=1) - Fills tensor with values drawn from a normal distribution with the given mean and std

• nninit.constant(tensor, val) - Fills tensor with the constant val

• nninit.xavier_uniform(tensor, gain=1) - Fills tensor with values according to the method described in "Understanding the difficulty of training deep feedforward neural networks" - Glorot, X. and Bengio, Y., using a uniform distribution.

• nninit.xavier_normal(tensor, gain=1) - Fills tensor with values according to the method described in "Understanding the difficulty of training deep feedforward neural networks" - Glorot, X. and Bengio, Y., using a normal distribution.

• nninit.kaiming_uniform(tensor, gain=1) - Fills tensor with values according to the method described in "Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification" - He, K. et al. using a uniform distribution.

• nninit.kaiming_normal(tensor, gain=1) - Fills tensor with values according to the method described in "Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification" - He, K. et al. using a normal distribution.

• nninit.orthogonal(tensor, gain=1) - Fills the tensor with a (semi) orthogonal matrix. Reference: "Exact solutions to the nonlinear dynamics of learning in deep linear neural networks" - Saxe, A. et al.

• nninit.sparse(tensor, sparsity, std=0.01) - Fills the 2D tensor as a sparse matrix, where the non-zero elements will be drawn from a normal distribution with mean=0 and std=std.