Week5 的题目被学友反馈是 frustrating。

笔者也是一路摸索过来的，把踩过的坑记录如下。

ML:Neural Networks: Learning

/learn/machine-learning/resources/EcbzQ

FAQ for Week 5 and programming exercise 4

/learn/machine-learning/discussions/weeks/5/threads/ag_zHUGDEeaXnBKVQldqyw

Unit Test

All Unit Test

/learn/machine-learning/discussions/all/threads/0SxufTSrEeWPACIACw4G5w

Ex4 Test Cases

ex4 test case for sigmoidGradient()Test cases for ex4 nnCostFunction()

Programming Exercise Test Cases (list) - 详细函数讲解

/learn/machine-learning/discussions/all/threads/m0ZdvjSrEeWddiIAC9pDDA

求 nnCostFunction的cost function，要理解下面两篇文章：

这里的技巧是要for循环。

ex4 Tutorial for forward propagation and cost

/learn/machine-learning/programming/AiHgN/neural-network-learning/discussions/threads/QFnrpQckEeWv5yIAC00Eog

Computing the NN cost J using the matrix product. 这里介绍了，为什么用for循环，为啥举证解决不了这里的二维求和函数。

/learn/machine-learning/discussions/all/threads/AzIrrO7wEeaV3gonaJwAFA

ex4 tutorial for nnCostFunction and backpropagation 实现逆传播，技巧是

不要用for循环，不要用for循环。要对着举证的行列数，挨个对比自己的实现。并且要用sigmoidGradient(z2)， 错误的方式是用方法sigmoid.

/learn/machine-learning/discussions/all/threads/a8Kce_WxEeS16yIACyoj1Q