Can not achieve 95% accuracy

Course Q&A TensorFlow Developer Professional Convolutional Neural Networks in TensorFlow
1

I have changed the architecture of the model many times but can’t get more than 91% accuracy!

2

Does this help?

3

I have changed:

  1. the architecture
  2. the conv size
  3. the dense size
  4. the optimizer
  5. the learning rate

I think I am missing some thing.
The accuracy reached 91% and the val accuracy 90%

4

Please click my name and message your notebook as an attachment.

5

I sent as attachment

6

Here are 2 hints:

  1. As far as conv filters go, increase the number of units as you go deeper in the network.
  2. Try adam optimizer.
1 Like
7

Thanks!
I added some layers and changed to adam and achieved it!

1 Like
8

@kisax
Your optimizer learning rate is too high. The default learning rate for Adam is .001 as shown in this link. Your optimizer is learning at the rate of .01

In general, start with a few learning rates in close proximity to the default learning rate when picking the best learning rate for your problem. My advice is to spend time on model architecture before adjusting the learning rate. Although learning rate is the most important hyperparameter for a NN, the defaults that ship with the framework work well for a wide variety of problems.

9

Thank you! It worked!