LEC 3
Mathematical Preliminaries
Common Discrete Probability Distributions
1. Bernoulli distribution
: 伯努利分布
models binary outcomes (coin flip). 模型二进制结果
P
(
X
=
head
) =
p
and
P
(
X
=
tail
) = 1 −
p
2.
Generalised Bernoulli distribution
: 广义伯努利分布
models
k
> 2
outcomes (rolls of a
k
sided die)
3. Binomial distribution: 二项式分布
models a sequence of multiple flip 模拟一个硬币的多次翻转的序列
4. Multinomial distribution: 多项分布
models a sequence of multiple rolls of a -sided die for k>2
If there are rolls and is the number of times the die came up on side , then the probability of this event is
Missing value
解决缺失值的方法：
1.discard
2.
fill in values by hand
3. set “missingValue”
4.
replace with the mean 用两个数之间的平均值来代替这个缺失的值
但是如果出现了一个outlier，那么就会很不准确
5. predict
We can train a new classifier to first predict the missing values in data instances and then train a second classifier to predict the target class using all (original + missing values predicted) the data points.
6.
accept missing values
Noisy value
Over-fitting vs. Under-fitting
过拟合 over-fitting：用于训练集的效果太好，而测试机的效果不好
Feature Normalisation
方法二：高斯归一化
