Machine learning Note (ii)-from Andrew Ng's instructional video

Omit the use of octave end, later use to see it

Week Three:

Logistic Regression:

For 0-1 categories

Hypothesis representation:

: Sigmoid function or Logistic function

Decision Boundary:

Theta's Transpose * small x>=0 is boundary

May:non-linear decision boundaries, constructing the polynomial of X

Cost function:

Simplified cost function and gradient descent:

Because Y has only two values, merging:

To find the least biased guide:

(The denominator should be ignored)

Advanced Optimization:

Conjugate gradient,bfgs,l-bfgs (Inquiry learning)

Multi-Class Classification:one-vs-all:

Once the logistic regression classification is used for each class, after determining the parameters, the category of Max is calculated: Called One-vs-all (a one-to-many method).

Regularization:the Problem of Overfitting

Overfiting:reduce number of features or regularization

Linear regression:

Gradient Descent:

Normal equation:

regularized Logistic regression:like linear Regression,add extra in J (theta)

Attention: Many regular items start from 1 and do not punish for 0.

Machine learning Note (ii)-from Andrew Ng's instructional video