Can not achieve desired accuracy

I have been trying to complete TF1 C2W4 assignment and could never get the accuracy past 4%. I have gone through most of the posts and tweaked the submission as follows:

  1. Noted that the rescale doesn’t happen twice
  2. Noted that I’ve larger number of features as we go deeper in the NN (increased 32->64)
  3. Have Dense layer with 256 activation functions
  4. Use Adam optimizer with learning rate 0.001
  5. Use sparse_categorial_crossentropy loss function

Despite this, I still have accuracy of 4% as shown below

“loss: nan - accuracy: 0.0410 - val_loss: nan - val_accuracy: 0.0462”

Much appreciate any insight.

Feedback after reviewing your notebook:

Please read this:

Welcome to this assignment! In this exercise, you will get a chance to work on a multi-class classification problem. You will be using the Sign Language MNIST dataset, which contains 28x28 images of hands depicting the 26 letters of the english alphabet.

Your NN has 3 output units. Please fix this.

Thanks @balaji.ambresh somehow I went in autopilot mode thinking it was rock-paper-scissors problem (even the pictures shown in intermediate stages did not wake me up). Will try and mark this as solution (I’ve run out of free gpu time on colab)