[Deep Learning Diary] Day 1: Hello world, Hello CNN MNIST

活动地址：CSDN21天学习挑战赛

2022年08月01日 星期一 天气晴

It's an unusual Monday,How many times have you stood upFlagAbandoned on Monday and on Tuesday,So I made up my mind once again with an oath：Continue to update the text to record the growth bit by bit,从周一开始！This time, the daily study record will be completed in the form of a diary,这是第一篇,也是新的开始！

严正声明：There is no sloppy eating behavior in this diary,Only based on the author's personal experience,如有雷同,纯属巧合！

创作计划

机缘：Record the growth of deep learning Xiaobai
预期：至少 21 天/The output of this article shares deep learning related notes
憧憬：Able to keep writing

时间	主题	内容	进度
2022-08-01	Hello world,Hello CNN MNIST!	Deep learning cloud environment experience、Learn about deep learning frameworks	已完成
2022-08-02	深度学习,“粮草”先行	Learn how to obtain datasets、Learn how deep learning framework datasets work	待编写
2022-08-03	Study late at night,只为“卷”–卷积神经网络入门详解	Identify by weather、Take clothing image classification as an example,Explain Convolutional Neural Networks	待学习

Cloud environment experience

The author believes that one of the obstacles to getting started with deep learning is that there is no corresponding environment to experience it,就连 AnimeGAN 作者Asher Chan They all complained that the limited computing power led to the delay of the paper,Of course there is a difference between doing research and just learning,After all we are just learning words,The free computing power provided by some cloud vendors is enough for us to run some cases.

为了避免争议,The following content is only relevant to the author, Just based on the author's personal experience,There is no promotion or derogation of any cloud vendor.如有不适,欢迎批评指正！

​The author probably came into contact with artificial intelligence-related development platforms and frameworks from the beginning of the epidemic,恰逢国内 Big4 Proposed by public cloud vendors”普惠AI“策略落地,I really feel it”Pratt & Whitney“of warmth–免费学习、Free computing power and even give away various gifts.In fact, both domestic and foreign cloud vendors,对于AIDevelopers have certain support plans,”注册就送￥$",This is not a game ad,It is one of the promotion methods of various cloud manufacturers.For deep learning novices,Maybe we really don't have local computing power,But there is a devout learning heart,怎么办？Then we will go to various platforms to prostitute computing power！The platforms below are ranked in no particular order

Colab

If you have conditional access to the external network,You can use it to your heart's content Colab, Colab 是 Colaboratory（合作实验室）的简称,借助 Colab We can write and execute in the browser Python 代码,并且无需任何配置、免费使用 GPU,It's also easy to share.Colab Provides an interactive environment for writing and executing code,称为 Colab 笔记本,Colab Notebook is made by Colab 托管的 Jupyter 笔记本.So we don't need to build a development environment locally,It is completely possible to use free resources in the cloud.按照 CSDN User feedback,Basically get it for free K80 的GPU,偶尔狗屎运也会有两次T4,But of course there are limits,B almost on aboutColab免费GPUThere are still some flaws,关于 Colab Please refer to the official documentation for specific problems encountered in use《Colaboratory常见问题解答》.
接着我们在 Colab A must-knock for getting started with deep learning Hello World 案例 – MNIST手写数字识别
新建hello_CNN_MNIST.ipynb,切换到 GPU 模式,编写代码,大致效果如下图：

完整代码请参考卷积神经网络（CNN）实现mnist手写数字识别, The current case is in Colab 平台运行结果如下：

当然,It is also very convenient to share with other small partners for collaboration：

是不是感觉很方便？From environment construction to model training to case sharing,A set of punches can be easily done in less than ten minutes！
Next, let's take a look at the domestic light,I think it's the closest to the original Jupyter 的 CodelLab.

CodeLab

作为ModelArts Developer Groups Guangzhou 的核心组织者,笔者对ModelArts 的 CodeLab 还是有一定的了解的,almost watching CodeLab Bit by bit, it has been updated and iterated to this relatively complete version,关于 CodeLab Please listen to Xiaosheng.CodeLab 是ModelArts推出的新功能,At present, it can only be used in Beijing Four,Free computing power specifications are preset with one-click access to the development environment,For committed to build“AI界Github”的AIGallery社区发布的Notebook样例（.ipynb格式文件）,可直接在CodeLab中打开,Check out the sample code shared by others.This is the same as mentioned above Colab有些类似.既然 CodeLab这么强大,We can't wait to see the beauty,Please let the author reveal it for youCodeLab的神秘面纱,近距离感受“普惠AI”的“真香”！

除了免费的 CPU 资源,还有免费的32G P100可供选择,Based on the author's long-term use CodeLabexperience：最高可获得8小时+的使用限额（PS：Here I look forward to the big guys who can write automatic continuation scripts）.
接着我们在Colab 平台导出hello_CNN_MNIST.ipynb并导出到CodeLab,Start our training journey.

I use the default one Python 3.6 + TensorFlow 1.13.1 环境,So the code needs simple modification,You can also click to experience一键Run in ModelArts

After the run is over, we run through completely CNN 实现 MNIST 手写数字识别：

TIANCHI

As a domestic veteran cloud manufacturer,The free resources given are also very powerful, CPU 无限使用,默认给到 60 小时GPU的额度,Points arrive200再送30小时,It can be used to learn temporarily to relieve urgent needs.官方介绍：“天池notebook集成机器学习PAI DSW（DataScienceWorkshop）Explorer Edition,Become the base of Tianchi Laboratory,Complete for everyoneIDEand abundant computing resources”,Let's take a look at the general interface first：

in TianchiNotebook中,基于Jupyter做了定制化开发,整个界面焕然一新,The following figure is the result of Tianchi operation：

同样的,本案例使用 CPU Training is also possible,It just takes a little more time：

结语

In fact, the deep learning environment in the cloud is far more than these,As long as you are good at discovering、勇于分享,We can get more free GPU 资源,We believe deep learning will become easier to learn,become connectedPython Xiaobai can also climb,Be able to train your own deep learning model and apply it to your life.
Thanks for reading this far,If you have any comments or suggestions on the article,欢迎在评论区留言互动！