Probability Basics:
## 随机变量的概率分布,随机变量的函数的概率分布,条件概率,联合概率
## 期望,方差,协方差,相关系数
## 里面有各种分布:
### 典型离散分布:
0-1分布:取值不是0就是1,两个概率都是50%, 只实验一次
伯努利分布: 也是只有两种结果,但是概率不一定是50%,所以0-1分布是伯努利分布的一种特例,只实验一次(也叫伯努利实验)
二项分布: 伯努利实验 做n次,有k成功的概率, 是一个二项式, n次实验之间是独立的
泊松分布: 一段时间内某个事件出现k次的概率,比如一天内发生火灾的次数,交通事故的次数,
泊松分布可作为二项分布的极限而得到。也就是n取无限大的时候
### 典型的连续分布:
正态
## 分布函数F和概率密度f:
对于连续随机变量 F(x)是指 随机变量X取值<=x的概率
f(x) 表示X取x的概率密度,一般需要通过积分求得分布函数, 要求概率密度,也是通过分布函数进行求导(微分)得到
Parameter Estimation:
Estimate the parameters of the overall distribution function based on the sampled samples: Note that the estimated target object is the parameter in the distribution
## Point estimation:
maximum likelihood estimation: Find a parameter, based on the existing samples, that maximizes the probability of generating these samples from this distribution jointly ---Transform into the problem of finding extreme values: two methods: 1. Find the solution of the partial derivative equation system (it must be derivable), 2. Gradient descent (does not require derivation, but it may fall into a local optimum, from In terms of performance, it is best to use stochastic gradient descent)
moment estimation: I still don’t understand
it
.
hypothetical test:
According to the sampling sample, to verify the hypothesis that the overall distribution belongs to a certain distribution, or the overall distribution is known, assuming that a certain parameter is a certain value, it is necessary to make a final decision based on the sample, whether to accept this hypothesis, or consider this hypothesis Is it reasonable?
There are various testing methods.
The specific theory is not very clear
, but we can check it with reference to various examples, and we can look up the table.
Theorem of Large Numbers, Central Limit Theorem:
When multiple independent variables act together, they generally obey the normal distribution (also called Gaussian distribution).
When the number of samples increases, the expectation is approximately equal to the mean