当前位置:网站首页>特征预处理
特征预处理
2022-08-05 04:08:00 【米卡粒】
2.4.1 什么事特征预处理
为什么要进行归一化、标准化
无量纲化
2.4.2归一化:对于归一化来说,如果出现异常点,影响了最大值和最小值,那么结果显然是会发生改变的
通过对原始数据进行变换把数据映射到(默认为[0,1])之间
异常值:最大值 最小值
2.4.3 标准化:对于标准化来说,如果出现异常点,由于具有一定数据量,少量的异常点,对于平均值的影响并不大,从而方差改变较小
(x - mean)/ std
标准差:集中程度
def minmax_demo():
"""
归一化
:return:
"""
# 1 获取数据
data = pd.read_csv("lizi")
data = data.iloc[:, :3]
print("data:\n", data)
# 2 实例化一个转换器类
# transfer = MinMaxScaler()
transfer = MinMaxScaler(feature_range=[2,3])
# 3 调用fit_transform
data_new = transfer.fit_transform(data)
print("data_new:\n", data_new)
return None
def stand_demo():
"""
标准化
:return:
"""
# 1 获取数据
data = pd.read_csv("lizi")
data = data.iloc[:, :3]
print("data:\n", data)
# 2 实例化一个转换器类
transfer = StandardScaler
# 3 使用fit_transfer
data_new = transfer.fit_transform(data)
print("data_new:\n", data_new)
return None
由于 factor_returns.csv 没有找到,所以不知道可不可以运行
def variance_demo():
"""
过滤低方差特征
:return:
"""
# 1、获取数据
data = pd.read_csv("factor_returns.csv")
data = data.iloc[:, 1:-2]
print("data:\n", data)
# 2、实例化一个转化器
transfer = VarianceThreshold(threshold=10)
# 3、调用fit_transform
data_new = transfer.fit_transform(data)
print("data_new:\n", data_new, data_new.shape)
return None
2.5.1 降维 - 降低维度
ndarray
维数 : 嵌套的层数
二维数组
此处的降维:降低特征的个数
效果: 特征与特征之间不相关
2.5.1 降维
特征选择
Filter过滤式
方差选择法:低方差特征过滤
相关系数 - 特征与特征之间的相关程度
取值范围: -1 ~1
特征与特征之间的相关性很高:
1)选取其中一个
2)加权求和
3)主成分分析
Embeded嵌入式
def variance_demo():
"""
过滤低方差特征
:return:
"""
# 1、获取数据
data = pd.read_csv("factor_returns.csv")
data = data.iloc[:, 1:-2]
print("data:\n", data)
# 2、实例化一个转化器
transfer = VarianceThreshold(threshold=10)
# 3、调用fit_transform
data_new = transfer.fit_transform(data)
print("data_new:\n", data_new, data_new.shape)
# 计算两个变量之间的相关系数
r1 = pearsonr(data["pe_ratio"], data["pb_ratio"])
print("相关系数:\n", r1)
r2 = pearsonr(data['revenue'], data['total_expense'])
print("revenue与total_expense之间的相关性:\n", r2)
return None
决策树 正则化 深度学习
主成分分析:
2.6.1 什么是主成分分析(PCA)
sklearn.decomposition.PCA(n_compinents=None)
n_components
def pca_demo():
"""
PCA
:return:
"""
data = [[2,8,4,5], [6,3,0,8], [5,4,9,1]]
# 1 实例化一个转换器类
transfer = PCA(n_components=2)
# 调用fit_transform(data)
data_new = transfer.fit_transform(data)
print("data_new:\n", data_new)
return None
小数 表示保留百分之多少的信息量
整数 减少到多少特征
2.6.2案例探究用户对物品类别的喜好细分
边栏推荐
- How to discover a valuable GameFi?
- 多御安全浏览器 V10.8.3.1 版正式发布,优化多项内容
- Bosses, I noticed that a mysql CDC connector parameters scan. The incremental. Sna
- 数组常用方法总结
- [Paper Notes] MapReduce: Simplified Data Processing on Large Clusters
- Hard power or soft power, which is more important to testers?
- iMedicalLIS listener (2)
- How to solve complex distribution and ledger problems?
- flink读取mongodb数据源
- Walter talked little knowledge | "remote passthrough" that something
猜你喜欢
bytebuffer use demo
[BJDCTF2020]EasySearch
炎炎夏日教你利用小米智能家居配件+树莓派4接入Apple HomeKit
Defect detection (image processing part)
[MRCTF2020]Ezpop(详解)
bytebuffer 内部结构
Dive into how it works together by simulating Vite
bytebuffer internal structure
token, jwt, oauth2, session parsing
[Software testing] unittest framework for automated testing
随机推荐
Shell script: for loop and the while loop
bytebuffer put flip compact clear 方法演示
UE4 通过互动(键盘按键)开门
Android 面试题——如何徒手写一个非阻塞线程安全队列 ConcurrentLinkedQueue?
Redis key basic commands
Walter talked little knowledge | "remote passthrough" that something
Hard power or soft power, which is more important to testers?
MySql index learning and use; (I think it is detailed enough)
35岁的软件测试工程师,月薪不足2W,辞职又怕找不到工作,该何去何从?
Android interview question - how to write with his hands a non-blocking thread safe queue ConcurrentLinkedQueue?
2022软件测试工程师最全面试题
UE4 为子弹蓝图添加声音和粒子效果
[GYCTF2020]EasyThinking
In the WebView page of the UI automation test App, the processing method when the search bar has no search button
关于#SQL#的迭代、父子结构查询问题,如何解决?
Ffmpeg - sources analysis
Mysql的undo log详解
Ali's local life's single-quarter revenue is 10.6 billion, Da Wenyu's revenue is 7.2 billion, and Cainiao's revenue is 12.1 billion
[SWPU2019]Web1
Qixi Festival earn badges