C2-W2 Excercise 8

Course Q&A Deep Learning Specialization Improving Deep Neural Networks: Hyperparameter tun
1

I get the results:
Original learning rate: 0.5
Updated learning rate after 10 epochs: 0.4854368932038835
Updated learning rate after 100 epochs: 0.3846153846153846

However the expected results are:
Original learning rate: 0.5
Updated learning rate after 10 epochs: 0.5
Updated learning rate after 100 epochs: 0.3846153846153846

How come the results after 10 epochs don’t change? They seem they should change. For the record I am uisng the following line to calculate learning_rate:

{moderator edit - solution code removed}

1 Like
2

You missed one subtlety in the mathematical formula that they gave you: the symbols \lfloor and \rfloor mean the mathematical “floor” function.

3

Thank you, although I am not really sure what is the purpose of the floor function here but it worked.
Thanks.

1 Like
4

If you think about the goal of the calculation, it is to change the learning rate not on every epoch but only periodically every time_interval epochs. So think about what that formula would do if you omit the “floor” function: the fraction would change on every epoch, right?