机器学习笔记
Random Variables 随机变量
Discrete Random Variables 离散随机变量
-
Bernoulli distribution伯努利(二项)分布
X = {0, 1} pmf p(x) = Bern(x|q), where p(1) = Bern(1|q) = Pr[x = 1] = q (we also write p(x = 1)) p(0) = Bern(0|q) = Pr[x = 0] = 1 1 q (we also write p(x = 0))
-
Categorical distribution 类别分布
X = {0, 1, ..., C-1} pmf x ∼ p(x) = Cat(x|q), where probability vector q = [q0, q1, ..., qCC 1] with qk = q0 + q1 + ... + qCC 1 = 1 p(k) = Cat(k|q) = Pr[x = k] = qk (we also write p(x = k))
one-hot vector
Continuous Random Variables 连续随机变量
Gaussian variable 高斯变量
MATLAB 代码
Description of any MATLAB function: help < namefunction >
Generate a Bernoulli rv Bern(q): binornd(1, q) or (rand(1) < q)
Generate N Bernoulli rvs Bern(q): binornd(1, q, N, 1) or (rand(N, 1) < q)
Generate a categorical rv Cat(q): mnrnd(1, q)
Generate N categorical rvs Cat(q): mnrnd(1, q, N) (produces one-hot vectors as rows)
Generate a Gaussian rv N (m,s2): normrnd(m, s) or (m + s ∗ randn(1))
Gaussian pdf N (x|m,s2): normpdf(x, m, s)
Plot a function y = f (x): plot(x, y)
Expectation期望
-
Bernoulli rv 伯努利分布:
-
categorical rv 类别分布:
-
gaussion 高斯分布
Ex=x^2 +σ^2
方差Variance
-
Bernoulli RV伯努利变量
Var=q(1-q) -
Gaussian RV 高斯变量
Var(x)=σ^2 -
categorical rv 类别变量
Linear Algebra 线性代数
L * 1 vector
-
一个x向量 L * 1 vector
-
transpose of vector x 转置矩阵
-
normalize a vector 正则化矩阵(求方向)
-
measures the cosine of the angle θ between vectors x and y 两个向量的夹角
L * M matrix 矩阵
-
a L * M matrix
-
transpose matrix 转置矩阵
-
diagonal matrix 对角矩阵 I L I_L IL 两种写法
Random Vectors 随机向量
- 一堆随机数的混合体
联合分布
-
Jointly Bernoulli random vector
-
Jointly Gaussian random vector
其中,协方差能体现两个向量的关系。P=1,两变量fully, and positively, correlated正相关。P=-1,两变量fully, and negatively, correlated 负相关。
Marginal Distributions 边缘分布
指在概率论和统计学的多维随机变量中,只包含其中部分变量的概率分布
- 离散型的例子
- 连续随机变量例子
Conditional Distributions 条件分布
- 离散型
P ( x 1 ∣ x 2 ) = P ( x 1 , x 2 ) P ( x 2 ) \boxed{P(x_1 | x_2) =\frac {P(x_1 , x_2)} { P( x_2)} } P(x1∣x2)=P(x2)P(x1,x2)
举个例子
- 连续型
例子:
Independence 独立
对于drv来说,
如果发现 X i X_i Xi的概率和 X L X_L XL无关, P ( X i ∣ X L ) = P ( X i ) \boxed{P(X_i | X_L)=P(X_i)} P(Xi∣XL)=P(Xi) ,那么就是独立
对于crv来说,协方差Cov=0则代表独立
Chain Rule连锁法则
没看懂,直接例题
Bayes Theorem贝叶斯定理
posterior distribution=prior distribution * likelihood ratio
Law of Iterated Expectations 迭代期望法则
名词解释
PMF : 概率质量函数probability mass function
PDF: 概率密度函数 probability density function
Var:方差variance
eigenvalues 特征向量
Cov:协方差covariance
discrete:离散的 discrete
subset:子集
待解决疑问
one-hot vector ?
categorical rv 类别变量的方差
p45
Cov? Pr? p56
p79 所有连续?
p81 ?
p94 BERN ?