Validation accuracy stuck at 20%

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

I’m working on an image classification project using this dataset. However, my validation accuracy is constantly stuck at exactly 20%.

I’ve tried changing epochs, optimizers, adding/decreasing layer complexity, batch_size, etc. I have a function that changes learning rate as the epochs change and I’m using data augmentation. I’m not sure what to do because my val_accuracy is not changing at all from 0.2. Any advice is greatly appreciated!

2

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

3

Please go through the lectures and previous labs in the course so far. Here are some issues:

  1. For a multiclass classification, sigmoid isn’t the correct activation to use for the output layer.
  2. ImageDataGenerator encodes labels as integers and doesn’t one-hot encode them. The way you’ve done it is incorrect. Do fix the loss function to reflect the labels.
  3. Your learning rate schedule picks a very low rate for the optimizer beyond 10 iterations. I recommend removing the callback and let the optimizer learn with the default configuration.
  4. Leave the batch size at 32 since you have a lot of images.
1 Like
4

Hello. Thank you for your help! I was able to correct these issues and my accuracy jumped to around 40%. Since this is still pretty low, I was wondering if you had any thoughts on what the foundational issues are with my program. I used a pre-existing cleaned dataset and I’ve tried a number of different epochs, optimizers, and loss functions to see if they make a difference. I’ve looked through the past lectures, and I’m still having trouble identifying what the core issue is. Do you have any thoughts on this?

Thank you so much for your help!

5

Posting notebooks in public is likely to get your account suspended / removed. Don’t forget to remove code from your post.

Please read this.
If you’re stuck after reading the provided topic, please click my name and message your notebook as an attachment.