Tuning hidden layer size to check accuracy

I am confused as to why the accuracy decreases after 5 units.

The model overfits after 5 hiddent units, but since we did not split into training and tests sets yet and the model is fit on all the data shouldn’t the accuracy continue to go up as more hidden units are added?

Maybe we have stopped too early? What happen if you increase num_iterations by 5 times? You might also set print_cost to True so that you can keep track of the cost change.


Thank you that worked. 20 units went up to 93% the number of iterations was causing the problem