# Data Science Libraries in Python

# toolboxes/libraries

# NumPy

introduces objects for multidimensional arrays, vectors and matrices, as well as functions that allow to easily perform advanced mathematical and statistical operations on those objects
介绍多维数组、向量和矩阵的对象，以及允许轻松对这些对象执行高级数学和统计操作的函数
provides vectorization of mathematical operations on arrays and matrices which significantly improves the performance
对数组和矩阵的数学运算进行矢量化，从而显著提高性能
many other python libraries are built on NumPy
许多其他 python 库都是基于 NumPy 构建的

# Pandas

adds data structures (data frame) and tools designed to work with table-like data
添加数据结构（数据框架）和用于处理类似表格的数据的工具
provides tools for data manipulation: reshaping, merging, sorting, slicing, aggregation etc.
提供数据处理工具：重塑、合并、排序、切片、聚合等。
allows handling missing data
允许处理丢失的数据

# SciPy

collection of algorithms for linear algebra, differential equations, numerical integration, optimization, statistics and more
线性代数、微分方程、数值积分、优化、统计学等算法的集合
part of SciPy Stack
SciPy 堆栈的一部分
built on NumPy
建立在 NumPy 之上
SciPy and NumPy are usually used for matrix-based operations, such as matrix factorization
SciPy 和 NumPy 通常用于基于矩阵的操作，例如矩阵分解

# SciKit-Learn

provides machine learning algorithms: classification, regression, clustering, model validation etc.
提供机器学习算法：分类、回归、聚类、模型验证等。
built on NumPy, SciPy and matplotlib
基于 NumPy、SciPy 和 matplotlib 构建

# Visualization libraries

# matplotlib

python 2D plotting library which produces publication quality figures in a variety of hardcopy formats
python 2D 绘图库，以各种硬拷贝格式生成出版物质量的图形
a set of functionalities similar to those of MATLAB
类似于 MATLAB 的功能
line plots, scatter plots, barcharts, histograms, pie charts etc.
折线图、散点图、条形图、直方图、饼图等。
relatively low-level; some effort needed to create advanced visualization
相对低级；创建高级可视化需要一些努力

# Seaborn

based on matplotlib
provides high level interface for drawing attractive statistical graphics
提供高级界面，用于绘制有吸引力的统计图形
Similar (in style) to the popular ggplot2 library in R
(样式上) 与 R 中流行的 ggplot2 库相似

# Pandas

# Introductions

Pandas is a newer package built on top of NumPy, and provides an efficient implementation of a DataFrame.
Pandas 是一个构建在 NumPy 之上的更新包，它提供了数据帧的高效实现。

DataFrames are essentially multidimensional arrays with attached row and column labels, and often with heterogeneous types and/or missing data.
DataFrame 本质上是具有附加的行和列标签的多维数组，并且通常具有异构类型和 / 或缺少数据。

# Basic data structure

# Series

Series is a one-dimensional array of indexed data.
序列是索引数据的一维数组。
The index-value is similar to key-value in dict in Python3.
We can create series by given a dict or an array/list
索引值类似于 python3 中 dict 的键值。

	print('Data Structure: Series')

	import numpy as np
	import pandas as pd

	# create series from an array ####################################

	# create Series from an array with automatic indices
	# 从具有自动索引的数组创建序列
	data = pd.Series([0.25, 0.5, 0.75, 1.0])
	print(data)
	print(data.values)
	print(data.index)
	print(data[1:4])

	# create Series from an array with given indices
	# 从具有给定索引的数组创建序列
	data = pd.Series([0.25, 0.5, 0.75, 1.0], index=['a', 'b', 'c', 'd'])
	print(data)
	print(data.values)
	print(data.index)
	print(data[1:4])
	print(data['c'])

	# it is not necessary to be continuous indices
	# 不一定要是连续的指数
	data = pd.Series([0.25, 0.5, 0.75, 1.0], index=[2, 5, 3, 7])
	print(data)
	print(data.values)
	print(data.index)

	# create series from a dict
	# 从字典创建序列
	population_dict = {'California': 38332521,
	'Texas': 26448193,
	'New York': 19651127,
	'Florida': 19552860,
	'Illinois': 12882135}
	data = pd.Series(population_dict)
	print(data)
	print(data.values)
	print(data.index)


	A = pd.Series([2, 4, 6], index=[0, 1, 2])
	B = pd.Series([1, 3, 5], index=[1, 2, 3])
	print(A)
	print(B)
	print(A + B)

	C=A.add(B, fill_value=0)
	print(C)

# DataFrame

DataFrame is a generalized array or a specialization of dict.
DataFrame 是广义上的数组或特殊的 dict。
It can be viewed as a table which stores data in different data types.
可以将其视为以不同数据类型存储数据的表。

	print('\nData Structure: DataFrame')

	population_dict = {'California': 38332521,
	'Texas': 26448193,
	'New York': 19651127,
	'Florida': 19552860,
	'Illinois': 12882135}
	population = pd.Series(population_dict)

	area_dict = {'California': 423967, 'Texas': 695662, 'New York': 141297, 'Florida': 170312, 'Illinois': 149995}
	area = pd.Series(area_dict)

	direction_dict={'California': 'West', 'Texas': 'Middle', 'New York': 'East', 'Florida': 'East', 'Illinois': 'Middle'}
	direction=pd.Series(direction_dict)

	# create a dataFrame by fusing three Series or dicts
	# 通过融合三个序列或 dict 创建 dataFrame
	states = pd.DataFrame({'population': population, 'area': area, 'direction': direction})
	print(states)
	print(states.index)
	print(states['area'])
	print(states[1:3])

	# create DataFrame in different ways: single column, multiple columns, given index and columns
	# 以不同的方式创建 DataFrame：单列、多列、给定索引和列
	df=pd.DataFrame(population, columns=['population'])
	print(df)
	df=pd.DataFrame({'population': population, 'area': area, 'direction': direction})
	print(df)
	df=pd.DataFrame(np.random.rand(3, 2), columns=['foo', 'bar'], index=['a', 'b', 'c'])
	print(df)

	# create dataFrame from dict
	# 从 dict 创建 dataFrame
	df=pd.DataFrame([{'a': 1, 'b': 2}, {'b': 3, 'c': 4}])
	print(df)

# Index

Index is the object associated with Series and DataFrame
索引是与 Series 和 DataFrame 关联的对象
It can be viewed as an immutable array (i.e., cannot be modified) or as an ordered set
可以将其视为不可变数组 (即不能修改) 或有序集

	indA = pd.Index([1, 3, 5, 7, 9])
	indB = pd.Index([2, 3, 5, 7, 11])

	# any set opeartions can be applied to index
	# 任何集合运算都可以应用于索引
	ind = indA & indB # intersection
	print(ind)
	ind = indA \| indB # union
	print(ind)
	ind = indA ^ indB # symmetric difference
	print(ind)

# Slicing data in dataFrame

loc and iloc are used to slice rows by default.
默认情况下， loc 和 iloc 用于切片行。
To slice columns, we can use df.loc[:, [1, 2, 3]]
要切片列，可以使用 df.loc[:, [1, 2, 3]]
we can use index -1 in iloc , but cannot use it in loc .
We only use index number in iloc , but we can use both index number and label in loc
ix is not suggested to be used in pandas v0.2 or later version

	# select data by columns
	# 按列选择数据
	print(states['area'])
	print(states[['area','direction']])

	# select data by rows
	# 按行选择数据
	print(states[1:3]) # index from [1, 3)
	print(states['Florida':'Illinois'])

	# select data by rows and columns
	# 按行和列选择数据
	print(states.iloc[1:3, 1:3]) # use numerical index in .iloc
	print(states.loc[['Texas','New York'], ['area','direction']]) # use string key in .loc

# `NaN` , `None` and mising values in Pandas

None is a general missing data
NaN can be interpreted as missing numerical data in float type
可以解释为浮点类型中缺少数值数据

	s=pd.Series([1, np.nan, 2, None])
	print(s)

	# check whether there are missing values
	# 检查是否有缺失值
	print(s.isnull())

	# drop missing values
	# 删除缺失值
	print(s.dropna())

	# in dataFrame, it will drop rows with null values
	# 在 dataFrame 中删除值为 null 的行
	df = pd.DataFrame([[1, np.nan, 2],
	[2, 3, 5],
	[np.nan, 4, 6]])
	print(df)

	# it will drop rows with null values
	# 删除值为 null 的行
	print(df.dropna())

	# it will drop columns with null values
	# 删除值为 null 的列
	print(df.dropna(axis='columns')) # axis =0 cols, axis =1 rows


	# fill in missing values
	# 填写缺失值
	data = pd.Series([1, np.nan, 2, None, 3], index=list('abcde'))
	print(data)
	print(data.fillna(0))

# File operations by Pandas

	import pandas as pd

	# load csv into dataFrame
	df1=pd.read_csv('Data_students.csv')
	print(type(df1))

	df2=pd.read_table('Data_students.csv',sep=',')
	print(type(df2))

	# show head and tail data
	print(df1.head(2))
	print(df1.tail(3))
	# print(df2.head(2))

	# get column names
	print(df1.columns)
	print(type(df1.columns)) # by default, they are index
	# convert column names to list
	col_list = df1.columns.tolist()
	print(col_list)
	print(type(col_list)) # it is list now

	# select data by rows
	print('\nselect data by rows')
	print(df1[1:3]) # get rows with index [1, 3)

	# select data by columns
	print(df1.columns)
	print('\nselect data by columns')
	cols = ['Age','Gender','Grade']
	print(df1[cols])

	# get mean value in a column
	print(df1['Age'].mean())

# Data Preprocessing

# Introductions

Data preprocessing may include the following operations:
数据预处理可能包括以下操作：

file load
文件加载
deal with missing values
处理缺失值
slicing data
切片数据
data normalization
数据标准化
data smoothing
数据平滑
data transformation, numerical to categorical
数据转换，从数字到分类
data transformation, categorical to numerical
数据转换，分类到数字
feature selection
特征选择
feature deduction
特征演绎
some special preprocessing, such as the operations in text mining, e.g., stopword removal, tokenization, TF-IDF weighting
一些特殊的预处理，例如文本挖掘中的操作，例如，停止词删除，标记化，tf-idf 加权

the following operations will use Data_Students.csv as the data set

# deal with missing values

	#Import Python Libraries
	import numpy as np
	import scipy as sp
	import pandas as pd
	import matplotlib as mpl
	import seaborn as sns
	from IPython.display import display, HTML

	df=pd.read_csv('data_students.csv')
	# get header
	cols=df.columns
	# get dimensions
	print(df.shape)

	# print header and dataType, as well as boolean value which tells missing values
	# 打印标题和数据类型，以及是否为缺失值的布尔值
	print(df.dtypes)

	print('ColumnName, DataType, MissingValues')
	for i in cols:
	print(i, ',', df[i].dtype,',',df[i].isnull().any())

	# print out and display dataframe as tables in HTML
	# 在 HTML 中打印并显示 dataframe 为表格
	display(HTML(df.head(10).to_html()))

	# By using GradeLetter as label, and visualize data
	# 使用 GradeLetter 作为标签，将数据可视化
	sns.set()
	sns.pairplot(df, hue='GradeLetter', height=2);

	# calculate mean value by ignoring missing values
	# 通过忽略缺失值来计算平均值
	mean_age=df['Age'].mean(skipna=True)
	mean_hr_assignment=df['Hours on Assignments'].mean(skipna=True)
	mean_hr_game=df['Hours on Games'].mean(skipna=True)
	mean_exam=df['Exam'].mean(skipna=True)
	mean_grade=df['Grade'].mean(skipna=True)

	# replace missing values in numerical variables by using mean value
	# 用均值代替数值变量中的缺失值
	df["Age"].fillna(df["Age"].mean(), inplace=True)
	df["Hours on Assignments"].fillna(df["Hours on Assignments"].mean(), inplace=True)
	df["Hours on Games"].fillna(df["Hours on Games"].mean(), inplace=True)
	df["Exam"].fillna(df["Exam"].mean(), inplace=True)
	df["Grade"].fillna(df["Grade"].mean(), inplace=True)

	# check again whether there are missing values
	# 再次检查是否有缺失值
	print('ColumnName, DataType, MissingValues')
	for i in cols:
	print(i, ',', df[i].dtype,',',df[i].isnull().any())

	# print out and display dataframe as tables in HTML
	# 在 HTML 中打印并显示 dataframe 为表格
	display(HTML(df.head(10).to_html()))

# Normalization

	# find numeric columns
	# 查找数字列
	numerics = ['int16', 'int32', 'int64', 'float16', 'float32', 'float64']
	# get column names
	# 获取列名
	cols_numeric = df.select_dtypes(include=numerics).columns.tolist()
	# get column indices
	# 获取列索引
	cols_numeric_index=[df.columns.get_loc(col) for col in cols_numeric]
	print('Numerical column names:\n',cols_numeric)
	print('Numerical column indeices:\n',cols_numeric_index)

	for i in cols:
	print(i, ',', df[i].dtype,',',df[i].isnull().any())

	# create a copy first
	# 先创建一个副本
	# 深复制，不会影响原始的 dataframe
	df_norm=df.copy(deep=True)

	# Normalization method 1 = z-score normalization
	# 归一化方法 1：z-score 归一化
	from sklearn.preprocessing import MinMaxScaler
	scaler = MinMaxScaler()
	df[cols_numeric]=scaler.fit_transform(df[cols_numeric])
	display(HTML(df.head(10).to_html()))


	# Normalization method 2 = Min-max normalization [0,1]
	# 归一化方法 2：最小最大归一化 [0,1]
	for col in cols_numeric:
	df_norm[col]=(df[col]-df[col].min())/(df[col].max()-df[col].min())

	# drop column ID since it is not useful in data science tasks
	# 删除 ID 列，因为它在数据科学任务中没有用
	# 1 = column 1 指的是列，如果是 0，则会在行 raw index 中搜索 ID
	df_norm=df_norm.drop('ID',1)

	df_norm.head(10)

# Data transformation

	df_transform=df_norm.copy(deep=True)
	# print out and display dataframe as tables in HTML
	# 在 HTML 中打印并显示 dataframe 为表格
	display(HTML(df_transform.head(5).to_html()))

	# convert numerical to categorical data, e.g., Age
	# 将数值转换为分类数据，例如年龄
	# 使用 cut 分成 8 组
	# df_transform['Age'] = pd.cut(df_transform['Age'],8)
	# display(HTML(df_transform.head(5).to_html()))

	# convert categorical data to numerical data, e.g., Degree
	# 将分类数据转换为数值，例如学位
	print(df_transform['Degree'].dtype)
	print(df_transform['Nationality'].dtype)

	df_dummies_degree=pd.get_dummies(df_transform['Degree'])
	print(df_dummies_degree.head(5))
	df_dummies_nation=pd.get_dummies(df_transform['Nationality'])

	# add binary variables to dataframe
	# 将二进制变量添加到 dataframe
	df_transform=df_transform.join(df_dummies_degree)
	df_transform=df_transform.join(df_dummies_nation)
	# remove the original categorical variable
	# 删除原始的分类变量
	df_transform=df_transform.drop('Degree',1)
	df_transform=df_transform.drop('Nationality',1)
	display(HTML(df_transform.head(5).to_html()))

	# N-1 binary variable is enough, drop 1
	# N-1 二进制变量就足够了，去掉 1
	df_transform=df_transform.drop(' PHD',1)
	df_transform=df_transform.drop(' China',1)

	display(HTML(df_transform.head(5).to_html()))

# Feature selection

	import matplotlib.pyplot as plt

	# print out and display dataframe as tables in HTML
	# 在 HTML 中打印并显示 dataframe 为表格
	display(HTML(df_transform.head(10).to_html()))

	# set features and labels
	# 设置特征和标签
	x = df_transform.drop('GradeLetter', 1)
	y = df_transform['GradeLetter']

	# Feature selection by using Filter model
	# 利用过滤模型进行特征选择

	# by using Pearson correlation as selecting criterion
	# 以皮尔逊相关系数作为选择标准
	# Pearson correlation can only be applied among numerical variables
	# 皮尔逊相关只能在数值变量之间应用
	# in this data, GradeLetter is highly correlated with numerical variable Grade
	# 在此数据中，GradeLetter 与数值变量 Grade 高度相关

	# calculate correlation and show in heatmap
	# 计算相关性并在热图中显示
	plt.figure(figsize=(12,10))
	cor = df_transform.corr()
	sns.heatmap(cor, annot=True, cmap=plt.cm.Reds)
	plt.show()

	# Correlation with output variable
	# 与输出变量的相关性
	cor_target = abs(cor["Grade"])
	# Selecting highly correlated features
	# 选择高度相关的特征
	relevant_features = cor_target[cor_target>0.5]
	print('\nSelected features by Filter model:\n',relevant_features)


	# Feature selection by using Wrapper model
	# 使用包装模型进行特征选择
	# A machine learning task is invovled in the Wrapper model
	# 包装器模型中包含了一个机器学习任务
	# We use the performance of the machine learning task to select influential features
	# 我们使用机器学习任务的性能来选择有影响力的特征
	# In this example, we use backward elimination in linear regression which predicts Grade
	# 在这个例子中，我们在预测 Grade 成绩的线性回归中使用反向消除法

	# # Backward Elimination 反向消除
	# import statsmodels.api as sm
	# cols = list(df_transform.columns)
	# cols.remove('GradeLetter') # drop the nominal variable
	# print('\n x variables: ',cols)

	# y=list(df_transform['Grade']) # using Grade as y variable in linear regression
	# pmax = 1
	# while (len(cols)>0):
	# p= []
	# X_1 = df_transform[cols]
	# X_1 = sm.add_constant(X_1)
	# model = sm.OLS(y,X_1).fit()
	# p = pd.Series(model.pvalues.values[1:],index = cols)
	# pmax = max(p)
	# feature_with_p_max = p.idxmax()
	# if(pmax>0.05):
	# cols.remove(feature_with_p_max)
	# else:
	# break
	# selected_features_BE = cols
	# print('\nSelected features by Wrapper model (regression):\n',selected_features_BE)


	# Feature selection by using Wrapper model
	# 使用包装模型进行特征选择
	# This example shows that we can use impurity criterion in decision trees to select important features
	# 这个例子表明，我们可以使用决策树中的杂质准则来选择重要的特征
	from sklearn.ensemble import ExtraTreesClassifier

	y = df_transform['GradeLetter']
	x = df_transform.drop('GradeLetter', 1)
	display(HTML(x.head(10).to_html()))

	model = ExtraTreesClassifier()
	model.fit(x, y)

	values=model.feature_importances_.tolist()
	keys=x.columns.tolist()
	d = dict(zip(keys, values))
	# sort pairs by values descending
	# 按值降序排列对
	s = [(k, d[k]) for k in sorted(d, key=d.get, reverse=True)]


	print('\nSelected features by Wrapper model (classification):\n')
	for k, v in s:
	print(k,'\t',v)

# Feature reduction

	# Example of PCA
	from sklearn.decomposition import PCA


	display(HTML(x.head(10).to_html()))
	# convert all features to numerical, in order to apply PCA
	# 将所有特征转换为数值，以便应用 PCA 主成分分析
	df_dummies_nationality=pd.get_dummies(df_transform['Nationality'])
	print(df_dummies_nationality.head(5))

	# add binary variables to dataframe
	# 将二进制变量添加到 dataframe
	df_transform=df_transform.join(df_dummies_nationality)
	# remove the original categorical variable
	# 删除原始的分类变量
	df_transform=df_transform.drop('Nationality',1)
	# remove on extra binary variable
	# 删除额外的二进制变量
	df_transform=df_transform.drop(' China',1)


	x = df_transform.drop('GradeLetter', 1)
	y = df_transform['GradeLetter']
	display(HTML(x.head(10).to_html()))

	# feature extraction 特征提取
	# API, https://scikit-learn.org/stable/modules/generated/sklearn.decomposition.PCA.html
	pca = PCA(n_components=10)
	fit = pca.fit(x)

	# summarize components 汇总组件
	# print("Explained Variance: %s") % fit.explained_variance_ratio_
	print('Explained variance: ', fit.explained_variance_ratio_)
	print('\nPCAs:\n', fit.components_)

	# select PCA and output new features
	# 选择 PCA 并输出新特征
	# for example, we choose the top-3 PCAs
	# 例如，我们选择排名前 3 位的 PCA

	PCAs = pca.fit_transform(x)
	PCAs_selected = PCAs[:,:3]
	df_PCAs = pd.DataFrame(data=PCAs_selected, columns=['PC1','PC2','PC3'])
	df_PCAs['GraderLetter']=y

	display(HTML(df_PCAs.head(10).to_html()))

	# write new data to external files
	# 将新数据写入外部文件
	df_PCAs.to_csv('Data_Students_PCA.csv', sep=',')

# Data Splits: Examples

	from sklearn.model_selection import train_test_split
	from sklearn.model_selection import cross_val_score
	from sklearn.model_selection import KFold

	from matplotlib import pyplot as plt
	import matplotlib as mpl
	import seaborn as sns

	print(df.columns)
	# Assume last column is your label, other columns are features
	# 假设最后一列是标签，其他列是特征
	X = df.loc[:, df.columns!='GradeLetter']
	y = df.loc[:,'GradeLetter']
	print(X.columns)
	print(type(X))
	print(type(y))


	# hold-out data split
	# API: https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.train_test_split.html
	x_train, x_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

	# N-fold data split
	# API: https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.KFold.html
	kf = KFold(n_splits=5, shuffle=True)
	data_5folds = []
	for train_index, test_index in kf.split(X,y):
	print("TRAIN:", train_index, "TEST:", test_index)
	# get actual data by index 按索引获取实际数据
	x_train, x_test = X.iloc[train_index], X.iloc[test_index]
	y_train, y_test = y[train_index], y[test_index]
	# save data into fold
	fold = [x_train, x_test, y_train, y_test]
	# add each fold data into 5folds
	data_5folds.append(fold)

	# N-fold cross validation
	# API: https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.cross_val_score.html
	# acc=cross_val_score(clf, x, y, cv=5, scoring='accuracy').mean()

数据挖掘机器学习

# Data Science Libraries in Python

# toolboxes/libraries

# NumPy

# Pandas

# SciPy

# SciKit-Learn

# Visualization libraries

# matplotlib

# Seaborn

# Pandas

# Introductions

# Basic data structure

# Series

# DataFrame

# Index

# Slicing data in dataFrame

# NaN , None and mising values in Pandas

# File operations by Pandas

# Data Preprocessing

# Introductions

# deal with missing values

# Normalization

# Data transformation

# Feature selection

# Feature reduction

# Data Splits: Examples

Midterm Review

Chapter 7 & 8

# `NaN` , `None` and mising values in Pandas