The distance from the point to the dividing surface

Interval of classifier (dataset)

Taking -1 and 1 facilitates mathematical processing

The value >=1 is a constraint condition

Need to maximize the interval

Difficulty solving

The aforementioned assumption is that the data is 100% linearly separable.

The constraints are changed to

any method

Numerical data

Helps to visualize separating hyperplanes

SVM spends most of its time training

Mainly through two parameter tuning

Simple calculation process

SVM can be used for almost any problem

It is a two-category classifier and needs to be modified for multiple categories.

Requires powerful computing power

Very complex to implement

Decompose a large optimization problem into multiple small optimization problems

short time

Find a series of alpha and b

Once these alphas are found, it is easy to calculate the weight vector w and obtain the separating hyperplane

Two alphas are selected for optimization in each loop

Once you find a suitable pair of alphas, increase one and decrease the other

Skipping the outer loop to determine the best alpha pair to optimize

less code

slow execution

Iterate over each alpha in the dataset

Randomly select another alpha from the remaining alpha set to form an alpha pair

Notice

Create an alpha vector and initialize it to a 0 vector

When the number of iterations is less than the maximum number of iterations (outer loop)

Only useful for NumPy types

alphas[alpha>0]

Two ways alternate

Maximum Uber length

In the end, only support vectors work

Think of it as wrapper/interface

Mapping from one feature space to another

good

How to replace inner product with kernel function

vector as argument

Ability to output a scalar based on vector distance operations



Too few: gets a poor decision boundary

Too many: nearly k nearest neighbors

A Comparison of Methods for Multiclass Support Vector Machines

Pattern Recognition