{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# NumPy概述\n",
"\n",
"\n",
"### NumPy简介\n",
"\n",
"<img src=\"images/ch8/1.png\" style=\"zoom:60%;\">\n",
"\n",
"NumPy(Numerical Python)是科学计算和数据分析的基础库，是进行数据处理与可视化的基础，也是科学Python生态系统的基础，支撑着几乎所有进行科学计算的 Python 库，包括 SciPy、Matplotlib、 pandas、 scikit-learn和 scikit-image等。\n",
"\n",
"|<img src=\"images/ch8/2.png\" style=\"zoom:100%;\">|<img src=\"images/ch8/3.png\" style=\"zoom:80%;\">|\n",
"|:----:|:----:|\n",
"\n",
"NumPy主要支持功能包括：\n",
"+ 提供快速、节约空间的多维数组，并具有矢量运算和复杂广播能力；\n",
"+ 为基于数组的计算提供了大量的标准数学函数；\n",
"+ 提供了线性代数、傅里叶变换、随机生成等高级数学功能；\n",
"+ 提供了C、C++、Fortran等语言的接口支持\n",
"\n",
"Python中的列表（List）可以保存一组数据，可以用来当作数组使用。但由于列表的元素可以是任何对象，因此列表中保存的是对象的指针，这种结构不适合做数值运算。Python也提供了array模块，能直接保存数值，但它不支持多维数组，也没有各种运算函数，也不适合做数值运算。NumPy的诞生弥补了这些缺陷，它提供了两种基本对象：\n",
"+ ndarray：n-dimensional array object，用于存储单一数据类型的多维数组。\n",
"+ ufunc：universal function object，能够能够对数组进行处理的标准函数。\n",
"\n",
"### NumPy导入\n",
"\n",
"通常推荐用以下的方式导入NumPy函数库，库名被缩写为np是一个被广泛采用的约定。\n",
"```python\n",
"import numpy as np\n",
"```\n",
"导入NumPy后，可使用NumPy库中相关帮助函数的获取相关文档与源码。\n",
"\n",
"+ **np.\\_\\_version\\_\\_**\n",
"\n",
"查看NumPy库的版本。"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import numpy as np\n",
"\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import numpy as np\n",
"\n",
"\n",
"np.__version__"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"+ **numpy.lookfor(what, module=None, import_modules=True, regenerate=False, output=None)**\n",
"\n",
"对NumPy文档执行关键字what搜索，返回搜索的结果。"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"np.lookfor('lookfor')\n",
"\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"+ **numpy.info(object=None, maxwidth=76, output=None, toplevel='numpy')**\n",
"\n",
"获取并返回函数、类或模块的帮助信息。"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"np.info(np.info) # 运行将返回numpy.info()函数相关文档"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"+ **numpy.source(object, output=<_io.TextIOWrapper>)**\n",
" \n",
"打印NumPy对象的源代码或将其写入文件。仅用 Python 编写的对象可返回源代码，库中许多用C语言定义的函数和类不会返回有用的信息。"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"np.source(np.source) # 运行将返回numpy.source()函数源代码"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"np.source(np.array) # numpy.array()函数由C语言实现，运行将返回'Not available for this object.'"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"学习NumPy可以帮助我们理解面向数组的编程思想，不过对于一般的数据分析与可视化而言，并不需要深入学习NumPy，掌握其中常用方法即可，如需相关具体信息可自行查阅[NumPy官方文档](https://numpy.org/devdocs/index.html)。\n",
"\n",
"### 数组的基本概念"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"与列表不同，**数组在创建时具有固定的大小，且数组中的所有元素的数据类型都要相同**。常见的包括一维数组、二维数组和三维数组。更高维度的数组使用比较少，此处不赘述。\n",
"<img src=\"images/ch8/4.png\" style=\"zoom:80%;\">\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"NumPy库axis表示轴的意思，**指定某个axis，就是沿着这个axis的方向做相关操作**。一维数组、二维数组和三维数组的axis见上图，明显可以发现，数组有几维就有几个轴。\n",
"\n",
"由于在数组中所有元素的数据类型都是一样的，所以数组的运算和存储效率相对于列表来说要高得多。而且，NumPy专门针对数组的操作和运算进行了设计，存储效率和输入输出性能远优于Python中的嵌套列表。通常数组越大，NumPy的优势就越明显。\n",
"\n",
"分别对有100000个数据的列表和数组，分别对其中每个数字求平方值。根据程序执行所消耗时间对比来比较其运行效率。"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import datetime # 导入datetime，用于计算程序运行时间\n",
"\n",
"start = datetime.datetime.now() # 记录程序开始时间\n",
"b = list(range(100000)) # 生成包含100000个数据的列表\n",
"a = [i**2 for i in b] # 使用列表推导式生成平方值列表\n",
"end = datetime.datetime.now() # 记录程序结束时间\n",
"during = (end - start).seconds * 1000 + (end - start).microseconds / 1000 # 计算耗时，精确到毫秒 \n",
"print(f'\\n耗时：{during}毫秒')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import datetime # 导入datetime，用于计算程序运行时间\n",
"import numpy as np\n",
"\n",
"start = datetime.datetime.now() # 记录程序开始时间\n",
"a = np.arange(100000) # 生成包含100000个数据的数组\n",
"b = a**2 # 使用数组运算生成平方值数组\n",
"end = datetime.datetime.now() # 记录程序结束时间\n",
"during = (end - start).seconds * 1000 + (end - start).microseconds / 1000 # 计算耗时，精确到毫秒 \n",
"print(f'\\n耗时：{during}毫秒')"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.7.5"
}
},
"nbformat": 4,
"nbformat_minor": 4
}
