tensorflow2.0与tensorflow1.0的性能区别介绍

ipython 

import mnist_data
mnist_data.save_mnist_mongodb()

0
100
200
300
400
500
...

Step 1600 : loss 0.597398758 ; accuracy 0.906712472
Step 1700 : loss 0.0526806675 ; accuracy 0.90900588
Step 1800 : loss 0.212036133 ; accuracy 0.911422193
Step 1900 : loss 0.245924264 ; accuracy 0.913889468
Step 2000 : loss 0.0638188794 ; accuracy 0.915765
20.71102285385132
Final step 2000 : loss tf.Tensor(0.06381888, shape=(), dtype=float32) ; accuracy tf.Tensor(0.915765, shape=(), dtype=float32)

pip install tensorflow==1.13.1

step 1700, training accuracy 0.960
step 1800, training accuracy 0.900
step 1900, training accuracy 0.930
12.46434211730957
test accuracy 0.942