我使用tensorflow 1.8，python 3.5和Windows 10来构建神经网络。 现在，我想收集一批输入作为神经网络的输入。但是我发现有些困惑。
当我将批次中的每个输入（例如分别包含3个数组）放入批次中时，我会得到如下结果：
input(1)
[[-0.02692897 -0.03908982 -0.01578733 -0.01504654 0.02450592 0.01947054
-0.02676752 -0.01917666 0.01936193 0.00408307 0.02334256 0.02143812
-0.0030923 -0.00462034 0.05150664 -0.00511832 -0.02197915 -0.00747188
0.00887368 0.02778551 0.02663407 -0.04100436 -0.01207645 0.02224508
-0.00502366 -0.01461473 0.05860103 0.00348206 -0.05176451 0.04181299
-0.00867486 -0.03567595 -0.05745618 0.05188627 0.00775612 0.01768746
0.00999217 -0.00296307 0.02664051 -0.04164632 0.05470607 -0.02736513
-0.01645982 -0.00916753 0.04894571 -0.02026801 0.0468831 0.02868102
-0.00475237 -0.02818991 -0.00358583 -0.01493043 0.0290164 -0.01078751
-0.0374933 0.01449602 0.04310944 -0.01696572 0.05669416 0.01611015
0.007667 -0.01522103 0.01240731 0.00879388 0.01272955 -0.03045401
-0.00846879 -0.01863868 -0.00854502 -0.05032277 0.01803133 0.02157342
-0.01348906 -0.04085188 -0.0191056 -0.014448 0.03539667 0.00082406
-0.06092907 0.00374502 -0.03160831 -0.02406421 0.03367792 -0.01895053
0.02847722 -0.01144491 0.02083589 0.00850848 0.00842468 0.03270741
-0.01254861 -0.03890391 -0.00217735 0.05016148 0.02249428 0.02832348
0.00260775 -0.01649905 0.01784565 -0.01581169]]
output
[[ 0.00396302 -0.01919217 -0.11362895]]
input(2)
[[-2.33860500e-03 -6.36590496e-02 -2.25870237e-02 -5.66520169e-02
-1.47972926e-02 2.16148910e-03 -2.86009517e-02 -2.54487991e-02
-5.74010285e-03 1.09514492e-02 1.41426474e-02 2.45856978e-02
-3.81153785e-02 -4.08029519e-02 7.04206973e-02 -3.96048138e-03
-1.49977617e-02 -2.57393196e-02 -1.28207020e-02 4.31949608e-02
2.56654564e-02 -8.19921307e-03 1.80648696e-02 -2.10555214e-02
-3.99227180e-02 5.55533171e-03 3.49735767e-02 -3.43401427e-03
-1.55919567e-02 6.03180118e-02 -2.19617505e-02 -4.63087037e-02
-5.51730543e-02 4.62662335e-03 -2.31043529e-02 -3.30487378e-02
1.08707463e-02 -2.39517204e-02 4.06590328e-02 -5.18664718e-02
4.03931886e-02 8.63777380e-03 -3.89002962e-03 5.23323797e-05
3.06941057e-03 -3.55914757e-02 2.48787254e-02 1.54354060e-02
-1.59684233e-02 3.14703174e-02 -2.86886226e-02 8.81753117e-03
1.56278200e-02 -3.60304825e-02 -6.47309870e-02 7.22086616e-03
2.72711534e-02 4.46651950e-02 4.75119315e-02 1.28488885e-02
3.04374984e-03 -2.88453922e-02 -7.26217264e-03 3.72415152e-03
6.41038129e-03 -1.08471129e-03 1.27645936e-02 -5.67848869e-02
-4.00970876e-02 -6.65567741e-02 1.21807251e-02 4.00495492e-02
-1.28227668e-02 -1.33121302e-02 -2.56709680e-02 4.54149162e-03
4.46238443e-02 -1.37233045e-02 -5.38579002e-02 1.78707577e-02
-8.11182894e-03 -1.50115415e-02 4.01578955e-02 -4.20019031e-02
4.06283997e-02 -7.22978190e-02 -2.08040774e-02 1.92635823e-02
2.46056560e-02 5.80835491e-02 -3.57985832e-02 -1.02872364e-02
-8.82912893e-03 3.30532677e-02 4.77855764e-02 -2.40224693e-03
-2.69014109e-02 -5.26522584e-02 1.47666584e-03 3.41499448e-02]]
output
[[-0.05926291 -0.10402986 -0.1300593 ]]
input(3)
[[-0.02692897 -0.03908982 -0.01578733 -0.01504654 0.02450592 0.01947054
-0.02676752 -0.01917666 0.01936193 0.00408307 0.02334256 0.02143812
-0.0030923 -0.00462034 0.05150664 -0.00511832 -0.02197915 -0.00747188
0.00887368 0.02778551 0.02663407 -0.04100436 -0.01207645 0.02224508
-0.00502366 -0.01461473 0.05860103 0.00348206 -0.05176451 0.04181299
-0.00867486 -0.03567595 -0.05745618 0.05188627 0.00775612 0.01768746
0.00999217 -0.00296307 0.02664051 -0.04164632 0.05470607 -0.02736513
-0.01645982 -0.00916753 0.04894571 -0.02026801 0.0468831 0.02868102
-0.00475237 -0.02818991 -0.00358583 -0.01493043 0.0290164 -0.01078751
-0.0374933 0.01449602 0.04310944 -0.01696572 0.05669416 0.01611015
0.007667 -0.01522103 0.01240731 0.00879388 0.01272955 -0.03045401
-0.00846879 -0.01863868 -0.00854502 -0.05032277 0.01803133 0.02157342
-0.01348906 -0.04085188 -0.0191056 -0.014448 0.03539667 0.00082406
-0.06092907 0.00374502 -0.03160831 -0.02406421 0.03367792 -0.01895053
0.02847722 -0.01144491 0.02083589 0.00850848 0.00842468 0.03270741
-0.01254861 -0.03890391 -0.00217735 0.05016148 0.02249428 0.02832348
0.00260775 -0.01649905 0.01784565 -0.01581169]]
output
[[ 0.00396302 -0.01919217 -0.11362895]]
但是如果我将这3个输入放入一个新数组中，并将这个新数组输入到神经网络作为输入，我会得到不同的结果：
input
[[-0.02692897 -0.03908982 -0.01578733 ... -0.01649905 0.01784565
-0.01581169]
[-0.00233861 -0.06365905 -0.02258702 ... -0.05265226 0.00147667
0.03414994]
[-0.02692897 -0.03908982 -0.01578733 ... -0.01649905 0.01784565
-0.01581169]]
output:
[[ 0.00396303 -0.01919215 -0.11362892]
[-0.05926289 -0.10402983 -0.1300593 ]
[ 0.00396303 -0.01919215 -0.11362892]]
您可以看到输出值略有变化。我认为这与tensorflow处理精度的方式有关，因为变化发生在最后两位。
但是有人知道我如何使批量输入的结果和单个输入的结果相同吗？