Bug小能手系列(python)_5:ValueError: Fan in and fan out can not be computed for tensor with fewer than 2

ValueError: Fan in and fan out can not be computed for tensor with fewer than 2

错误

在python开发中，使用xavier_normal_函数初始化报错。

# 初始化代码块
x = torch.tensor([1.0, 0.5, 0.3, 0.2])
torch.nn.init.xavier_normal_(x)
print(x.shape)

报错。提示

ValueError: Fan in and fan out can not be computed for tensor with fewer than 2

解决方案

该报错是xavier_normal_初始化用xavier_normal_初始化参数时候, 输入的参数维度至少是2, 当输入维度只有1维时报错。
这里不建议更改pytorch库的源代码.使用unsqueeze()进行维度扩展再输入到xavier_normal_进行初始化.
对应代码为：

x = torch.tensor([1.0, 0.5, 0.3, 0.2])
# 先进行维度扩展
torch.nn.init.xavier_normal_(x.unsqueeze(0))
print(x) # tensor([ 0.2140, -0.4359, -0.1479, -0.6696])