coursera machine learning andrew ng

Before the machine learning is very interested in the holiday cannot to see Coursera machine learning all the courses, collated notes in order to experience repeatedly.I. Introduction (Week 1)-What's machine learningThere is no un

is close to the global minimum. In fact, you can dynamically adjust the learning rate α= constant 1/(number of iterations + constant 2), so that as the iteration, α gradually reduced, in favor of the final convergence to the global minimum value. However, because "constant 1" and "Constant 2" is not OK, so often set α is fixed.How do you judge the convergence of the model as the iteration progresses? Every 1000 or 5,000 samples, the J value of these

Overview Cost Function and BackPropagation Cost Function BackPropagation algorithm BackPropagation Intuition Back propagation in practice Implementation Note:unrolling Parameters Gradient Check Random initialization Put It together Application of Neural Networks Autonomous Driving Review Log 2/10/2017:all the videos; Puzzled about Backprogation 2/11/2017:reviewed backpropaga

general, multiplication does not satisfy the exchange law: $ \ Matrix {A} \ times \ matrix {B} \ not = \ matrix {B} \ times \ matrix {A} $Special Matrix $ \ Matrix {I }=\ matrix {I _ {n \ times N }}=\ begin {bmatrix} 1 0 \ cdots 0 0 \ Cr0 1 \ cdots 0 0 \ Cr \ vdots \ vdots \ Cr0 0 \ cdots 1 0 \ Cr0 0 \ cdots 0 1 \ Cr \ end {bmatrix} $ For any matrix $ \ matrix {A} $: $ \ Matrix {A} \ times \ matrix {I }=\ matrix {I} \ times \ matrix {A }=\ matrix {A} $Inverse Matrix and inverte

m>=10n and uses multiple Gaussian distributions.In practical applications, the original model is more commonly used, the average person will manually add additional variables.If the σ matrix is found to be irreversible in practical applications, there are 2 possible reasons for this:1. The condition of M greater than N is not satisfied.2. There are redundant variables (at least 2 variables are exactly the same, XI=XJ,XK=XI+XJ). is actually caused by the linear correlation of the characteristic

different lambda, the calculated training error and cross-validation error are as follows:Lambda Train error Validation error 0.000000 0.173616 22.066602 0.001000 0.156653 18.597638 0.003000 0.190298 19.981503 0.010000 0.221975 16.969087 0.030000 0.281852 12.829003 0.100000 0.459318 7.587013 0.300000 0.921760 1.000000 2.076188 4.260625 3.000000 4.901351 3.822907 10.000000 16.092213 9.945508The graphic is represented as follows:As

Extremely light of a semester finally passed, summer vacation intends to learn the big step down this machine learning techniques.The first lesson is the introduction of SVM, although I have learned it before, but I heard a feeling is very rewarding. The blogger sums up a ballpark figure, and the specifics areTo listen: http://www.cnblogs.com/bourneli/p/4198839.htmlThe blogger sums it up in detail: http://w

)/∂ (θ (1) JK) is tested for gradients. After the partial derivative code does not have a problem, close the Gradient check section code.6. Use gradient descent or other advanced algorithms to perform reverse propagation to find the θ values for minimizing j (θ).This paper describes the gradient descent algorithm in neural networks: starting from the random initial point, descending step by step, until the local optimal value is obtained. Algorithms such as gradient descent can at least guarante

would the Vectorize this code to run without all for loops? Check all the Apply. A: v = A * x; B: v = Ax; C: V =x ' * A; D: v = SUM (A * x); Answer: A. v = a * x; v = ax:undefined function or variable ' Ax '. 4.Say you has a vectors v and Wwith 7 elements (i.e., they has dimensions 7x1). Consider the following code: z = 0; For i = 1:7 Z = z + V (i) * W (i) End Which of the following vectorizations correctly compute Z? Check all the Apply.

dimension.Finally, we propose a method for solving overfitting, including data cleaning/pruning, data hinting, regularization (regularization), confirmation (validation), andTo drive for example to illustrate the role of these methods, the latter two methods are also the contents of the following two lessons.Data cleaning/pruning is to correct or delete the wrong sample points, processing is simple, but usually such sample points are not easy to find.Data hinting generate more sample numbers by

This section is about regularization, in the optimization of the use of regularization, in class when the teacher a word, not too much explanation. After listening to this class,To understand the difference between a good university and a pheasant university. In short, this is a very rewarding lesson.First of all, we introduce the reason for regularization, simply say that the complex model with a simple model to express, as to how to say, there is a series of deduction hypothesis, very creative

-Feature ScalingWhen we are faced with multidimensional feature problems, we need to ensure that the multidimensional features have similar scales, which will help the gradient descent algorithm to converge faster.Take the housing price forecast problem as an example, assuming that the two characteristics we use, namely the size of the house and the number of rooms, the size value range is 0-2000 square feet, and the value of the room number is 0-5, which causes the gradient descent algorithm to

-Cost functionFor the training set and our assumptions, we will consider how to determine the coefficients in the assumptions.What we are going to do now is to choose the right parameters, and the selection of parameters directly affects the accuracy of the resulting straight line for the training set description. The difference between the predicted value and the actual value in the training set is the modeling error (Modeling error).the cost function is defined by calculating the sum of square

than or equal to 0, which is greater than or equal to 3 o'clock, the model predicts y = 1.We can draw a straight line, which is the dividing line of our model, separating the area predicted to 1 and the area predicted as 0.What kind of model would be appropriate if our data were to be presented in the following circumstances?Because curves are required to separate areas of y = 0 and y = 1, we need two-character:Assuming that the parameter is [-1 0 0 1 1], then we get the decision boundary is ex

, i.e., all of our training examples lie perfectly on some straigh T line. If J (θ0,θ1) =0, that means the line defined by the equation "y=θ0+θ1x" perfectly fits all of our data. For the To is true, we must has Y (i) =0 for every value of i=1,2,..., m. So long as any of our training examples lie on a straight line, we'll be able to findθ0 andθ1 so, J (θ0,θ1) =0. It is not a necessary that Y (i) =0 for all of our examples. We can perfectly predict the value o

-Learning RateIn the gradient descent algorithm, the number of iterations required for the algorithm convergence varies according to the model. Since we cannot predict in advance, we can plot the corresponding graphs of iteration times and cost functions to observe when the algorithm tends to converge.Of course, there are some ways to automatically detect convergence, for example, we compare the change value of a cost function with a predetermined thr