Loss Increasing


In Week 2, I trained and evaluated below networks using the code:
tf.keras.metrics.mean_absolute_error(x_valid, results).numpy()

  • single layer neural network : 4.9779134
  • DNN with given learning rate : 4.9461427 (loss improves)
  • DNN with optimized learning rate = 7e-6 at 100 epochs : 5.091236 (loss increases)
  • DNN with optimized learning rate = 7e-6 at 500 epochs : 6.0789175 (highest loss)

The mae increased the more I tuned the model, so I assumed an error with my code.
Then I ran the “Deep neural network notebook (Lab 3)” provided by the course.
Here, “DNN with optimized learning rate = 8e-6 at 500 epochs” also resulted in the highest mae.

Is this an error in the notebook? Or am I interpreting mae results incorrectly?

In this weeks assignment the loss metric is not given the mae!

I’m talking about Lab 3, not the ungraded assignment.
Lab 3 contains the code
“tf.keras.metrics.mean_absolute_error(x_valid, results).numpy()”
I believe that means it measure mae. The github/google colab link to lab below.

This is TF1 and this is the page for TF-AT, you are in the wrong course here.