1w[1],b[1]没变化

2w[2],b[2]两列变化一样

Zeros()

Ones()

Constant(value=0)

RandomNormal(mean=0.0, stddev=0.05, seed=None)

RandomUniform(minval=-0.05, maxval=0.05, seed=None)

TruncatedNormal(mean=0.0, stddev=0.05, seed=None)

VarianceScaling(scale=1.0, mode='fan_in', distribution='normal', seed=None)

Orthogonal(gain=1.0, seed=None)

Identity(gain=1.0)

glorot_normal(seed=None)

he_normal(seed=None)

lecun_normal(seed=None)

he_uniform(seed=None)

glorot_uniform(seed=None)

lecun_uniform(seed=None)

Zeros

Ones

Constant

RandomNormal

RandomUniform

RandomUniform

TruncatedNormal

VarianceScaling

scale: 缩放尺度（正浮点数）

mode:  "fan_in", "fan_out", "fan_avg"中的一个，用于计算标准差stddev的值。

distribution：分布类型，"normal"或“uniform"中的一个。

如果mode = "fan_in"， n为输入单元的结点数；

如果mode = "fan_out"，n为输出单元的结点数；

如果mode = "fan_avg",n为输入和输出单元结点数的平均值。

limit = sqrt(3 * scale / n)

Orthogonal

gain是最后矩阵乘以的系数

Identity

gain是乘以该矩阵的系数

[[ 1.,  0.,  0.],

[ 0.,  1.,  0.],

[ 0.,  0.,  1.]]

lecun_uniform

VarianceScaling(scale=1.,

mode='fan_in',

distribution='uniform',

seed=seed)

lecun_normal：

VarianceScaling(scale=1.,

mode='fan_in',

distribution='normal',

seed=seed)

he_normal

VarianceScaling(scale=2.,

mode='fan_in',

distribution='normal',

seed=seed)

he_uniform

VarianceScaling(scale=2.,

mode='fan_in',

distribution='uniform',

seed=seed)

glorot_normal

VarianceScaling(scale=1.,

mode='fan_avg',

distribution='normal',

seed=seed)

glorot_uniform

VarianceScaling(scale=1.,

mode='fan_avg',

distribution='uniform',

seed=seed)