Compilation takes forever

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

Whenever, I re-run the code, it takes forever to complete. I spent 3 days on C2W2 and most of my time was wasted on waiting for the code to finish compiling.
I tried to connect to GPU backend but was unable to. I got the message indicating that I need to upgrade to Colab Pro.
Is there a way to speed up the process without dropping the accuracy?

2

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

3

Thank you for messaging your notebook. There’s nothing in the notebook that should take you 3 days to run your code.

Google colab has a free tier. So, if you want a higher chance of getting a GPU, consider upgrading to a paid tier.

It’s ok to use your desktop / laptop to write this code as well. Be sure to use the same version of tensorflow as on the colab environment.

4

Thanks for your reply. I apologize for the miscommunication.
The 3 days duration accounts for the whole process of changing the code and re-running it. The time for running the code once is approximately a 1.5 hour to 2 hours.
The problem I am having is that I can’t reach the 80% accuracy. Whenever, I make changes and re-run the code it takes very long. It is becoming frustrating to complete the assignment. Do you have any hints, for example which parameter to look at?
Thanks in advance

5

Sure. Here are some hints:

  1. Please grab a machine with GPU for quicker results.
  2. As far as the architecture is concerned, start from a smaller NN and add layers / nodes when you don’t achieve the desired accuracy. See ungraded labs for the week for examples.
  3. Have number of conv filters as powers of 2 and increase them as you go deeper in the network.
  4. For optimizer, I recommend adam unless you are willing to spend time on tuning the learning rate.
7

Could you please clear my one doubt, so there are many optimizer and I am not sure when we should select Adam, RMSprop, Adadelta, Adagrad, and SGD. Could you please give me some suggestion which will be helpful for me during the Tensorflow certification exam.

8

Choice of optimizer and its learning rate are hyperparameters of a NN. Please see this for learning to tune a NN.