I did have some trouble completing this exercise.
When comparing equations 6.31 and 6.32 with what is in lxmls-toolkit/lxmls/deep_learning/mlp.py, I can't figure out if the code is correct or not:
when computing nabla_W and nabla_b: the code iterates through the geometry of the network, rather than the training set (like in equations 6.31/32)
in line 138 it uses $\tilde z^n$ instead of $x^m$ -- although I believe this is a typo in the guide.
The results seem to be good, so the code is likely correct, but I can't see why that would be equivalent.
Also, in equation 6.32, the sum doesn't seem to depend on $m$.
Just some small fixes on exercise 6.2.
I did have some trouble completing this exercise. When comparing equations 6.31 and 6.32 with what is in lxmls-toolkit/lxmls/deep_learning/mlp.py, I can't figure out if the code is correct or not:
The results seem to be good, so the code is likely correct, but I can't see why that would be equivalent. Also, in equation 6.32, the sum doesn't seem to depend on $m$.