高维
目录:
正文:
一、什么是高维问题


二、如果出现过拟合情况怎么办
第一种方法较为简单:pre-screening和step-wise或者best subset


第二种是常见的惩罚项:penalization

几种常见的惩罚方法:
- Ridge penalization

- LASSO

LASSO的缺点


- LARS


- Coordinate descent

- Boosting

- Bridge

group bridge

- SCAD



上面几种的补充:




- Marginally Differential Variables




- Bootstrap


- TGDR


三、变量选择的方法
1,leave one out, 2. resampling, 3. Inference-based

Model building with variable selection

① 模型选择与假设检验 ② best subset and selection criteria ③ Resampling methods
one 介绍; two 变量选择的传统方法;



three 变量选择 之 贝叶斯方法(Bayesian and stochastic search)
Bayesian model selection

spike and slab prior

stochastic search

fourth 变量选择之正则化
Nonnegative garrote

LASSO and bridge regression

stochastic search

其他正则化方法

fifth 继续前进
group variable selection
四、维度降减
PCA,PLS,TCA,STR

Lasso , group lasso

五、解决计算问题

- screening

- integrative analysis

- meta-analysis 多个数据集分析

六、system-based analysis


cluster network


七、Laplacian penalization

一个总结:




Markov random field :
① Gaussian Markov Random Field (GMRF)
先验


估计

② Binary Markov random field (BMRF)
先验

估计

③ scaled binary markov random field

variable selection and response prediction

上面这些算法的simulation 和 result

A simple regulatory network (RegN)
