“This is the 16th day of my participation in the Gwen Challenge in November. Check out the details: The Last Gwen Challenge in 2021.”
Introduction to the
The first three chapters have introduced the contents of Bayesian network structure in detail, and the structure learning and parameter learning of Bayes in detail, so that we can clearly know which aspects should be started in Bayesian learning, so that we can learn Bayes faster and simpler. Today I’m going to focus on the subsets of Bayes, static And dynamic bayes.
Static bayes
When researching and applying BN, the influence of time factor on system and data is not considered.
There are some minor differences between Bayesian networks and static Bayesian networks:
- Hypothesis premise:
Bayesian network: all variables are discrete; Each feature has dependence (uncertain causal reasoning) relation (variable independent); After the value of each node’s precursor node is determined, this node condition is independent of all its indirect precursor nodes. Bayesian networks relax the assumption that each variable is independent.
Static Bayes: the features are independent of each other; In static Bayes, for the problem that several conditional probability values do not exist, it is generally solved by adding all probability values, and the influence on the explained variables is consistent, so variable screening cannot be carried out.
- Should case:
Bayesian network: in the case of incomplete information, the relationship between adjacent words and synonyms in text classification can be solved by deducing unobservable random variables from observable random variables.
Static Bayes: Classification
- Disadvantages:
Bayesian networks: Variables cannot be filtered because the assumption of consistent influence on the explained variables cannot be relaxed.
Static Bayes: The establishment of static Bayes among features that are not independent from each other increases the complexity of the model.
- Advantages:
Bayesian network: The combination of Bayesian principle and graph theory builds a mathematical model based on probabilistic reasoning, which has strong advantages for solving complex uncertainty and relevance problems. Insensitive to missing data, you can learn about causality and deepen your understanding of the data. Can integrate prior knowledge into modeling; The problem of over-fitting is avoided and data need not be retained for testing.
Static Bayes: Simple. For the given item to be classified, the type with the largest conditional probability will be selected, which is the ideological basis of static Bayes.
Dynamic bayes
Bayesian network, also known as causal probability network, is the combination of graph theory and probability theory. It has the advantages of powerful knowledge reasoning, intuitive expression ability, clear topology structure and convenient decision-making mechanism, and mainly researches on data mining for uncertain problems.
The combination of static Bayes and time series constitutes dynamic Bayes (DBN). DBN is a new stochastic model which has the ability to deal with time series data. Compared with BN network, DBN has advantages in dealing with temporal nonlinear uncertainty problems. However, due to the complexity of dynamic Bayesian network, expert knowledge and BN structure learning algorithm are not suitable for DBN.
remarks
As long as this chapter introduces the difference between static Bayes and dynamic Bayes and Bayesian networks, we can better distinguish what is static Bayes and what is dynamic Bayes, and we can choose how to apply these two networks in our study and work to solve practical problems. The next chapter introduces some application prospects of Bayesian networks.
Past classic
- Bayes review (1) – Gold Digging (juejin. Cn)
- Bayes review (2) – Gold Digging (juejin. Cn)
- Bayes review (3) – Nuggets (juejin. Cn)