Robotic Learning Cornerstone (Machine learning foundations) Learn Cornerstone job Four after class exercise solution

Hello everyone, I am mac Jiang, today and you share the coursera-ntu-machine learning Cornerstone (Machines learning foundations)-job four of the exercise solution. I encountered a lot of difficulties in doing these topics, when I find the answer on the Internet but can not find, and Lin teacher does not provide answers, so I would like to do their own questions on how to think about the writing down, for everyone to provide some ideas. Of course, my understanding of the topic is not necessarily correct, if you bo friends found errors please contact, thank you! Again: Please do not use this blog as a way to pass the exam, or better to learn and understand the course! Hope my blog is helpful to your study!

The source of this article: http://blog.csdn.net/a1015553840/article/details/51173679

Other job analysis See summary stickers: http://blog.csdn.net/a1015553840/article/details/51085129

1. First question

(1) Test instructions: When using hyphothesis set H there is a deterministic noise (fixed noise), if a smaller hyphothesis set h ', then the fixed noise is increased or decreased?

(2) Analysis: First we need to know what is fixed noise

Fixed noise is caused by the fact that the target function f itself is too large for QF. If the QF of F itself is too large, then it is not easy to use H to fit this higher objective function, so the fixed noise is large. When we use the smaller h ' instead of H to fit F, because H ' is smaller, the fitting degree of f is even worse, so the deterministic noise will increase!

(3) Answer: Increasing second item

2. The second question

(1) Test instructions: Define H (q,c,q0), which one of the following is correct.

(2) Analysis: First, we need to understand the definition, and then bring into the calculation.

H (10,0,3) = {h (x) = [w0 W1 W2 0 0 0 0 0 0 0 0] * [z0 Z1 Z2 Z3 Z4 Z5 Z6 Z7 Z8 z9 Z10] = w0*z0 + w1*z1 + w2*z2}

H (10,0, 4) = {h (x) = [w0 W1 W2 W3 0 0 0 0 0 0 0] * [z0 Z1 Z2 Z3 Z4 Z5 Z6 Z7 Z8 z9 Z10] = w0*z0+w1*z1+w2*z2+w3*z3}

h2={h (x) = {W0*Z0+W1*Z1+W2*Z2}

H (10,0,3) AC h (10,0,4) = H2

(3) Answer: The second is the same as the other

3. Question Three

(1) Test instructions: Using decay as Regularizer, using the gradient descent method to calculate the minimum value of the Eaug, the iteration step is ITA, then each iteration update the formula?

(2) Analysis:

(3) Answer: the second

4. Question Fourth

(1) Test instructions: The monotonicity of the mode of Wreg (lambda) change with lambda

(2) Analysis

Using physical meaning analysis (1) When W ' * W = c does not include win, then increase the lambda, that is, reduce C, then even more can not include win, | | wreg| | = C Decrease

(2) When W ' * W =c includes win, | | wreg| | =|| win| |, then increase the lambda, that is, reduce C, if you still include win, then it will not change

It's not monocytogenes.

(3) Answer: non-increasing

5. Question Fifth

(1) Test instructions: Calculates the error rate of two hyphothesis by using the leave one out cross validation, calculates by the square error method, and asks them to error the value of the parameter Rou

(2) This problem Bo Master also will not, may be the topic understanding wrong, hope the big God answer!

6. Question Sixth

(1) Test instructions: The problem is about survivor bias.

(2) Analysis: For the first time to 32 people e-mail, half said team A will win, Half said B team will win, there must be half of the mail (16) is correct; the first time the results come out, to the first correct 16 people e-mail, half said a will win, Half said B will win, and so on. Then 32 people in 5 games may have a person 5 times to accept the answer is correct.

(3) Answer: The third, the second need for the first time the correct 16 people e-mail

7. Question Seventh

(1) Test instructions: According to the method of the 6th question, each letter needs 10 pieces, if the sixth game someone spends 1000 yuan, ask this liar to earn how much money

(2) Analysis: The total need to send a total of 630 pieces of 32+16+8+4+2+1=63 letter, earned 370 yuan

(3) Answer: 370

8. Title

(1) Test instructions: A bank originally issued a credit card with a formula A (x), and then the 10,000 people in these credit cards were used as a sample to analyze the issue of credit cards for new users. Before you look at these samples, you use mathematical theory to present a credit card issuance formula that asks the size of the hyphothesis set.

(2) Analysis: Do not think the problem is too complicated, because you use mathematical deduction to put forward a credit card issuance formula, that Hyphothesis has been determined, so hyphothesis set size of 1

(3) Answer: 1

9. The ninth question

(1) Test instructions: Using hoeffding bound to calculate the probability of the difference between Ein and eout is not more than 1%, when the training sample number is 100000

(2) Analysis:

Since the eighth question already knows that M = 1 is finite, we can call this formula:

(3) Answer: 0.271

10. Question Tenth

(1) Test instructions: You use 100,000 samples from the bank to get a good g, he can fit the training samples well. But when we use him for practical purposes, we find that his generalization ability is weak.

(2) Analysis: The samples that should be given to us are obtained from the formula eighth A (x), so our training data is not clean and is contaminated with a (x). That is, we use a (x) and g (X) to determine the final machine learning algorithm performance.

(3) Answer: A (x) and g (x)

11. Question 11th

(1) test instructions: Adding K samples to the original n samples, then using these n+k samples to calculate the linear regression, the formula of the parameter

(2) Answer: the second

12. Question 12th

(1) Test instructions: If the method of 11 questions is used, then when the 11 formula is equal to the solution of the regularization logistic regression

(2) Analysis:

The formula of the regularization logistic regression is WREG, so that the formula of 11 questions equals him, that is, the fifth item can be

(3) Answer: item Fifth

13-20 answers please see: http://blog.csdn.net/a1015553840/article/details/51173020

The source of this article: http://blog.csdn.net/a1015553840/article/details/51173679

Other job analysis See summary stickers: http://blog.csdn.net/a1015553840/article/details/51085129

Robotic Learning Cornerstone (Machine learning foundations) Learn Cornerstone job Four after class exercise solution