def Batchnorm_simple_for_train(x, gamma, beta, bn_param):
"""
param:x : 输入数据,设shape(B,L)
param:gama : 缩放因子 γ
param:beta : 平移因子 β
param:bn_param : batchnorm所需要的一些参数
eps : 接近0的数,防止分母出现0
momentum : 动量参数,一般为0.9, 0.99, 0.999
running_mean :滑动平均的方式计算新的均值,训练时计算,为测试数据做准备
running_var : 滑动平均的方式计算新的方差,训练时计算,为测试数据做准备
"""
running_mean = bn_param['running_mean'] #shape = [B]
running_var = bn_param['running_var'] #shape = [B]
results = 0. # 建立一个新的变量
x_mean=x.mean(axis=0) # 计算x的均值
x_var=x.var(axis=0) # 计算方差
x_normalized=(x-x_mean)/np.sqrt(x_var+eps) # 归一化
results = gamma * x_normalized + beta # 缩放平移
running_mean = momentum * running_mean + (1 - momentum) * x_mean
running_var = momentum * running_var + (1 - momentum) * x_var
#记录新的值
bn_param['running_mean'] = running_mean
bn_param['running_var'] = running_var
return results , bn_param
滑动平均法记录mean和var,用于测试。
def Batchnorm_simple_for_test(x, gamma, beta, bn_param):
"""
param:x : 输入数据,设shape(B,L)
param:gama : 缩放因子 γ
param:beta : 平移因子 β
param:bn_param : batchnorm所需要的一些参数
eps : 接近0的数,防止分母出现0
momentum : 动量参数,一般为0.9, 0.99, 0.999
running_mean :滑动平均的方式计算新的均值,训练时计算,为测试数据做准备
running_var : 滑动平均的方式计算新的方差,训练时计算,为测试数据做准备
"""
running_mean = bn_param['running_mean'] #shape = [B]
running_var = bn_param['running_var'] #shape = [B]
results = 0. # 建立一个新的变量
x_normalized=(x-running_mean )/np.sqrt(running_var +eps) # 归一化
results = gamma * x_normalized + beta # 缩放平移
return results , bn_param
【深度学习】批归一化(Batch Normalization):https://blog.csdn.net/vict_wang/article/details/88075861
Batch Normalization原理与python实现:https://zhuanlan.zhihu.com/p/100672008
基础 | batchnorm原理及代码详解:https://blog.csdn.net/qq_25737169/article/details/79048516