## 1. Setting up your Machine Learning Application

#### 1.1 Train/Dev/Test sets

Make sure that the dev and test sets come from the same distribution。

Not having a test set might be okay.(Only dev set.)

So having set up a train dev and test set will allow you to integrate more quickly. It will also allow you to more efficiently measure the bias and variance of your algorithm, so you can more efficiently select ways to improve your algorithm.

#### 1.2 Bias/Variance

High Bias: underfitting
High Variance: overfitting

Assumption——human: 0% (Optimal/Bayes error), train set and dev set are drawn from the same distribution.

Train set error Dev set error Result
1% 11% high variance
15% 16% high bias
15% 30% high bias and high variance
0.5% 1% low bias and low variance

#### 1.3 Basic Recipe for Machine Learning

High bias –> Bigger network, Training longer, Advanced optimization algorithms, Try different netword.

High variance –> More data, Try regularization, Find a more appropriate neural network architecture.

## 2. Regularizing your neural network

#### 2.1 Regularization

In logistic regression, $$w \in R^{n_x}, b \in R$$$$J(w, b) = \frac {1} {m} \sum _{i=1} ^m L(\hat y^{(i)}, y^{(i)}) + \frac {\lambda} {2m} ||w||_2^2$$$$||w||_2^2 = \sum _{j=1} ^{n_x} w_j^2 = w^Tw$$
This is called L2 regularization.

$$J(w, b) = \frac {1} {m} \sum _{i=1} ^m L(\hat y^{(i)}, y^{(i)}) + \frac {\lambda} {2m} ||w||_1$$
This is called L1 regularization. w will end up being sparse. $\lambda$ is called regularization parameter.

In neural network, the formula is $$J(w^{[1]},b^{[1]},…,w^{[L]},b^{[L]}) = \frac {1} {m} \sum _{i=1} ^m L(\hat y^{(i)}, y^{(i)}) + \frac {\lambda} {2m} \sum _{l=1}^L ||w^{[l]}||^2$$$$||w^{[l]}||^2 = \sum_{i=1}^{n^{[l-1]}}\sum _{j=1}^{n^{[l]}} (w_{ij}^{[l]})^2, w:(n^{[l-1]}, n^{[l]})$$

This matrix norm, it turns out is called the Frobenius Norm of the matrix, denoted with a F in the subscript.

L2 norm regularization is also called weight decay.

#### 2.2 Why regularization reduces overfitting?

If $\lambda$ is set too large, matrices W is set to be reasonabley close to zero, and it will zero out the impact of these hidden units. And that’s the case, then this much simplified neural network becomes a much smaller neural network. It will take you from overfitting to underfitting, but there is a just right case in the middle.

#### 2.3 Dropout regularization

Dropout will go through each of the layers of the network, and set some probability of eliminating a node in neural network. By far the most common implementation of dropouts today is inverted dropouts.

Inverted dropout, kp stands for keep-prob:

$$z^{[i + 1]} = w^{[i + 1]} a^{[i]} + b^{[i + 1]}$$$$a^{[i]} = a^{[i]} / kp$$

In test phase, we don’t use dropout and keep-prob.

#### 2.4 Understanding dropout

Why does dropout workd? Intuition: Can’t rely on any one feature, so have to spread out weights.

By spreading all the weights, this will tend to have an effect of shrinking the squared norm of the weights.

#### 2.5 Other regularization methods

• Data augmentation.
• Early stopping

## 3. Setting up your optimization problem

#### 3.1 Normalizing inputs

Normalizing inputs can speed up training. Normalizing inputs corresponds to two steps. The first is to subtract out or to zero out the mean. And then the second step is to normalize the variances.

If the network is very deeper, deep network suffer from the problems of vanishing or exploding gradients.

#### 3.3 Weight initialization for deep networks

If activation function is ReLU or tanh, w initialization is: $$w^{[l]} = np.random.randn(shape) * np.sqrt(\frac {2} {n^{[l-1]}}).$$ This is called Xavier initalization.

Another formula is $$w^{[l]} = np.random.randn(shape) * np.sqrt(\frac {2} {n^{[l-1]} + n^{[l]}}).$$

#### 3.4 Numberical approximation of gradients

In order to build up to gradient checking, you need to numerically approximate computatiions of gradients.

$$g(\theta) \approx \frac {f(\theta + \epsilon) - f(\theta - \epsilon)} {2 \epsilon}$$

Take matrix W, vector b and reshape them into vectors, and then concatenate them, you have a giant vector $\theta$. For each i:

$$d\theta _{approx}[i]= \frac {J(\theta_1,…,\theta_i + \epsilon,…)-J(\theta_1,…,\theta_i - \epsilon,…)} {2\epsilon} \approx d\theta_i=\frac {\partial J} {\partial \theta_i}$$

If $$\frac {||d\theta_{approx} - d\theta ||_2} {||d\theta_{approx}||_2 + ||\theta||_2} \approx 10^{-7}$$, that’s great. If $\approx 10^{-5}$, you need to do double check, if $\approx 10^{-5}$, there may be a bug.

#### 3.6 Gradient checking implementation notes

• Don’t use gradient check in training, only to debug.
• If algorithm fails gradient check, look at components to try to identify bug.
• Remember regularization.
• Doesn’t work with dropout.
• Run at random initialization; perhaps again after some training.