July algorithm--December machine Learning online Class-13th lesson notes-Bayesian network

July Algorithm--December machine Learning online Class -13th lesson notes-Bayesian network

July algorithm (julyedu.com) December machine Learning Online class study note http://www.julyedu.com

?

1.1 The thought of Bayesian formula: The given result pushes the cause;

1.2 Assumptions of Naive Bayes

1, probability of a characteristic occurrence, independent of other characteristics (conditions) (characteristic independence)

2, each feature is equally important

For example: Text classification, Word appears as 1, does not appear as 0

Bayesian formula:

Decomposition:

?

Laplace smoothing

Determine the distance between two documents: the angle cosine

Determine the correct rate of classifiers: cross-validation

If a word appears more often, one less, you can use the counting method (but the previous experience, the effect generally)

1.3 Bayesian Networks

Bayesian Network: A Direction-free graph model, a probabilistic graph model, which investigates a set of random variables and their n-group conditional probability distributions according to the topological structure of probabilistic graphs

There is a relationship between the two random variables, but not the causal relationship

?

1.3.1 Simple Bayesian Network

The upper and right graphs are equivalent.

1.3.2 Fully connected Bayesian network

Full connection diagram of 5 nodes

1.3.3 Normal Bayesian network

The above formula and the right figure are a meaning

1. X1 and X2 Independent
2. X6 and X7 are independent under x4 given conditions

?

?

?

?

?

Joint distribution of all random variables:

1.3.4 Special Bayesian Network

Nodes form a chain network called the Markov model.

Only with regard to, with ... Independent

Pseudo-Random generator:

?

1.3.5 by Bayesian Network to determine the conditions of independence (three cases)

1. First case (Tail-to-tail)

2. The second type (head-to-tail)

3. The Third Kind (head-to-head)

N-node, n-order Markov model: full-Join model

Summary: Do not accumulate the complexity of the model, using the Occam ' s Razor in a timely manner.

Construction of 1.3.6 Bayesian network

Build Model: NB,HMM

Discriminant Model: SVM,LOGISTIC,RF

An example is as follows:

?

?

July algorithm--December machine Learning online Class-13th lesson notes-Bayesian network