分类目录：《深入浅出Pytorch函数》总目录

---

torch.nn.init模块中的所有函数都用于初始化神经网络参数，因此它们都在torc.no_grad()模式下运行，autograd不会将其考虑在内。

根据Glorot, X.和Bengio, Y.在《Understanding the difficulty of training deep feedforward neural networks》中描述的方法，用一个均匀分布生成值，填充输入的张量或变量。结果张量中的值采样自$U(-a,a)$，其中：
$$a=\text{gain}\times \sqrt{\frac{6}{\text{fan\_in}+\text{fan\_put}}}$$

这种方法也被称为Glorot initialization。

语法

torch.nn.init.xavier_uniform_(tensor, gain=1)

参数

• tensor：[Tensor] 一个$N$维张量torch.Tensor
• gain ：[float] 可选的缩放因子

返回值

一个torch.Tensor

实例

w = torch.empty(3, 5)
nn.init.xavier_uniform_(w, gain=nn.init.calculate_gain('relu'))

函数实现

def xavier_uniform_(tensor: Tensor, gain: float = 1.) -> Tensor:r"""Fills the input `Tensor` with values according to the methoddescribed in `Understanding the difficulty of training deep feedforwardneural networks` - Glorot, X. & Bengio, Y. (2010), using a uniformdistribution. The resulting tensor will have values sampled from:math:`\mathcal{U}(-a, a)` where.. math::a = \text{gain} \times \sqrt{\frac{6}{\text{fan\_in} + \text{fan\_out}}}Also known as Glorot initialization.Args:tensor: an n-dimensional `torch.Tensor`gain: an optional scaling factorExamples:>>> w = torch.empty(3, 5)>>> nn.init.xavier_uniform_(w, gain=nn.init.calculate_gain('relu'))"""fan_in, fan_out = _calculate_fan_in_and_fan_out(tensor)std = gain * math.sqrt(2.0 / float(fan_in + fan_out))a = math.sqrt(3.0) * std  # Calculate uniform bounds from standard deviationreturn _no_grad_uniform_(tensor, -a, a)