tensorflow學習之tf.truncated_normal和tf.random_noraml的區別

tf版本1.13.1，CPU

最近在tf裏新學了一個函數，一查發現和tf.random_normal差很少，因而記錄一下。。

一、首先是tf.truncated_normal函數

tf.truncated_normal(shape, mean=0.0, stddev=1.0, dtype=tf.float32, seed=None, name=None) 

shape是張量維度，mean是正態分佈是均值，stddev是正態分佈的標準差；

它是從截斷的正態分佈中輸出隨機值，雖然一樣是輸出正態分佈，可是它生成的值是在距離均值兩個標準差範圍以內的，也就是說，在tf.truncated_normal中若是x的取值在區間（μ-2σ，μ+2σ）以外則從新進行選擇。這樣保證了生成的值都在均值附近。

二、接下去是

tf.random_normal(shape, mean=0.0, stddev=1.0, dtype=tf.float32, seed=None, name=None)

參數設置和上一個函數是同樣的；而tf.random_normal是沒有要求必定在（μ-2σ，μ+2σ）以內的

栗子以下：

r1 = tf.random_normal(shape=[10,10], mean=0, stddev=1)
r2 = tf.truncated_normal(shape=[10,10], mean=0, stddev=1)
with tf.Session() as sess:
    print(sess.run(r1))
    print('-----------------------------')
    print(sess.run(r2))


###
[[-0.9133466  -0.2193664   0.9871885   0.98474556  0.6795205   0.5485084
  -0.5377412  -0.74770564 -1.1065788  -0.6537805 ]
 [-0.05684219 -0.40211347  0.29256898  0.05646685 -0.09572198 -0.5329077
   0.09256991  0.5098056  -0.56273437  3.0135465 ]
 [-0.76685655  0.80393404  1.4993162  -0.97593653 -0.02311829  0.18108685
  -0.10103939 -0.1788411   0.70498335  0.7432045 ]
 [ 1.0753849  -1.552068   -0.82677794 -1.0526472   0.35524416  1.109456
  -0.46477854 -0.94029546  0.4074978  -0.6275004 ]
 [ 1.1464692  -0.7329785   1.3993146  -1.3458818  -1.0205774  -1.3928136
   0.31929192 -0.19847174  0.24104206 -0.30365643]
 [-0.56316936  0.35070175 -1.7900109   0.18886246  0.70969003  0.61488485
   1.0273159   0.08935942  0.31060082  0.46601346]
 [-2.2100964  -0.5587107   0.23800382  0.58263725  0.44721082  0.17937267
  -1.6707375  -0.69823587 -0.624195    0.7153361 ]
 [-0.8212442  -0.4100253  -0.29756552  1.6561007   0.22083926  0.97440094
  -0.08767799  0.07237837  1.5110539   1.7204924 ]
 [ 0.7634979  -0.38169494 -0.07168189 -1.0445783  -0.4177571  -0.06731904
   0.13163103  0.73196214 -0.32269892 -1.9275837 ]
 [ 0.73937386 -0.08055622  1.209047   -0.41532582  0.11017569  0.7899525
  -1.8012363  -0.84279794  1.6240916   0.73638594]]
-----------------------------
[[-0.14283383  1.0008699   1.5809501   0.82382834 -0.6668856   0.59398574
  -1.0554414   0.87018394  0.2878338  -0.4894875 ]
 [ 1.3494765  -0.13940284  0.38545245 -0.16343059  0.37798592 -1.119075
  -0.9302422  -1.1171802  -0.28318515  1.5838846 ]
 [ 0.01883612  1.0887331  -0.18458353 -0.48704016 -0.84166986  1.2319418
   0.6718625  -0.7222486   0.88431233 -0.31767374]
 [-0.3927616   0.42229542  0.29394206  0.636135   -0.9557136  -0.14583842
   0.09705613 -1.5379425   0.6139084   0.36891633]
 [-1.0034062  -0.7085579   0.05415478 -0.26299214 -1.4239995   0.24866848
  -0.08754523  0.9556532  -0.70344573  0.5501471 ]
 [-0.36907005 -0.18001153 -1.7696055  -0.8719723   0.00751158  1.4784805
   1.2351555   0.5596131   0.9836762   1.3182775 ]
 [-0.68398184 -0.74510956  1.5121812  -0.5380244  -0.2752701   0.03485487
   1.495663   -1.2325596   0.94225466 -1.386867  ]
 [-0.00377628 -0.13128701  0.02556802  1.1236848   1.139232   -0.53410244
   0.87148935  1.9706047   1.2066965   0.9876827 ]
 [-0.3875238  -0.06041284  0.44305998  1.193       0.6871842   0.2273079
   1.6827972   1.1394504  -1.4383765  -0.21280776]
 [ 1.5706499   1.6732877  -1.1468259   0.3528784   1.8091112   1.2832314
  -0.32502323  0.34072635  1.5236534   0.89802533]]

可見r1裏面有一個數是在（-2,2）範圍之外的，而r2則全在（-2,2）範圍內