我如何在20小时内为AWS ML专业课程做好准备并进行破解

I am a great fan of how Tesla is executing the problem of gathering data from the fleet of cars to train their net in efforts to build FSD, hence the image.

我非常热衷于特斯拉如何执行从车队收集数据以训练他们的网络以建立FSD并因此建立形象的问题。

Although Amazon recommends 1–2 years of experience developing, architecting, or running ML/deep learning workloads on the AWS Cloud, before sitting for the exam. But, with the right mindset for preparation, anyone can ace the exam in much less time.

尽管亚马逊建议参加考试之前，建议在AWS云上开发，架构或运行ML /深度学习工作负载具有1-2年的工作经验。 但是，只要有了正确的准备思想，任何人都可以在更少的时间内获得考试的成功。

你真的喜欢ML吗？ (Do you really like ML?)

If you don’t like Machine Learning, this story is not for you. But if you are new to ML and interested to learn, it will be useful.

如果您不喜欢机器学习，那么这个故事不适合您。 但是，如果您不熟悉ML并且有兴趣学习，它将很有用。

I believe it becomes easy to learn a tool or concept when you really know the purpose of its existence or the impact of its existence. Knowing the purpose makes it clear and simple for you to map while you learn.

我相信，只要您真正了解工具或概念的存在目的或存在的影响，就可以轻松地学习它。 知道目的后，您在学习时就可以清楚，轻松地进行映射。

If ML is entirely new to you? Go, do your homework on:

如果ML对您来说是全新的？ 快去做功课：

1. What is Machine Learning?
   什么是机器学习？
2. What is a Distribution?
   什么是分布？
3. What is a Model?
   什么是模型？
4. Layman's understanding of steps to build an ML model?
   Layman对构建ML模型的步骤的理解？
5. What was the bottleneck for ML Systems before Cloud?
   Cloud之前ML Systems的瓶颈是什么？
6. How the idea of Cloud, opened possibilities for ML?
   云的想法如何为ML打开了可能性？

If you think the above ones are basic and boring, please try to cook some motivation by researching these.

如果您认为以上内容是基本且无聊的，请尝试通过研究这些内容来激发一些动力。

1. How are folks at Tesla training their Autopilot?
   特斯拉的员工如何训练他们的自动驾驶仪？
2. What did it take to make a product like Apple Card?
   制造Apple Card这样的产品需要什么？
3. How does a Skydio Autonomous drone work?
   Skydio自主无人机如何工作？
4. How is Siri suggesting you apps based on your usage patterns?
   Siri如何根据您的使用方式向您推荐应用程序？
5. How recommendation engine of YouTube works?
   YouTube的推荐引擎如何工作？
6. How China ingeniously achieved mass surveillance using Facial Recognition?
   中国如何通过面部识别巧妙地实现了大规模监视？
7. How Apple FaceId works?
   Apple FaceId如何工作？
8. What happens once you ask your Alexa to switch on your bedroom light?
   一旦您要求Alexa打开卧室灯会怎样？
9. How are big tech companies trying to achieve Level 5 Autonomous Driving?
   大型科技公司如何努力实现5级自动驾驶？
10. If you are in your 20s, you will surely know the app Shazam. Did you ever think of how Shazam works?
    如果您20多岁，那么您肯定会知道该应用程序Shazam。 您是否想到过Shazam的工作方式？
11. How do your insurance companies decide your premium?
    您的保险公司如何决定您的保费？
12. How brilliant are the features of photo editing apps in 2020?
    2020年照片编辑应用程序的功能多么出色？
13. How tabbed inboxes in Gmail classify E-mails?
    Gmail中的选项卡式收件箱如何对电子邮件进行分类？
14. Ever heard of Amazon Go?
    听说过Amazon Go吗？
15. How Visual Search works on Amazon?
    视觉搜索如何在亚马逊上工作？
16. How Doordash / Uber Eats / Postmates are determining an ETA for your order?
    Doordash / Uber吃得如何/邮递员如何确定您的订单的预计到达时间？
17. How Google Photos curates a series of best photos from your albums
    Google相册如何策划您相册中的一系列最佳照片
18. Should I still continue with the list? The list just goes on.
    我应该继续列出吗？ 列表继续。

Hopefully, this made you realize how deeply ML is responsible for the basic experiences we indulge every day in 2020, or at least you knew all this time that ML is the reason for these to be made possible. All right! In that case, you already are halfway through this goal, all the following steps are just to-dos and let you achieve the goal.

希望，这使您意识到ML对我们在2020年每天沉迷的基本体验负有多深的责任，或者至少您一直以来都知道ML是使之成为现实的原因。 行！ 在这种情况下，您已经完成了这一目标，以下所有步骤只是要做的事情，可以让您实现目标。

您是否偶然知道AWS？ (Did you happen to know about AWS already?)

Super! If you already knew about AWS or any Cloud platform, you are good.

超！ 如果您已经了解AWS或任何云平台，那就太好了。

Otherwise, I would suggest you know about Cloud, and also spend a few minutes skimming the AWS home page and the ML services they provide.

否则，我建议您了解Cloud，并花一些时间浏览一下AWS主页及其提供的ML服务。

Here is a recent lecture by David J. Malan, which I deeply suggest for beginners. — Click here for the lecture

这是大卫·J·马兰(David J. Malan)最近的演讲，我向初学者强烈建议。 — 单击此处进行讲座

Also, an introductory video of AWS.

另外，还有AWS的入门视频。

I would suggest you take a course on the basics of AWS.

我建议您参加有关AWS基础知识的课程。

All you need to be aware of are the benefits of Cloud and how the AWS Console looks.

您只需要了解Cloud的优势以及AWS Console的外观即可。

AWS ML专业考试真正测试了您什么？ (What really AWS ML Specialty exam tests you?)

60% ML and 40% AWS.

60％ML和40％AWS。

Before we dive into the steps, let’s take a look at what the exam wants you to be good at?

在深入探讨这些步骤之前，让我们看看考试希望您擅长什么？

As Amazon Web Services quotes them:

正如Amazon Web Services引用的那样：

1. Select and justify the appropriate ML approach for a given business problem

1.针对给定的业务问题选择并证明适当的ML方法

2. Identify appropriate AWS services to implement ML solutions

2.确定适当的AWS服务以实施ML解决方案

3. Design and implement scalable, cost-optimized, reliable, and secure ML solutions

3.设计和实施可扩展，成本优化，可靠和安全的机器学习解决方案

Here, though only #1 talks about ML alone, without #1 it makes very little sense of #2 & #3. So it basically revolves around your understanding on the process of building an ML model — gathering the data, cleaning the data, normalizing the data, labeling the data (if required), ability to choose the model, most importantly tuning the model’s hyperparameters, deploying the model into production and about ensuring the performance in production.

在这里，尽管只有＃1单独谈论ML，但没有＃1的话，对＃2和＃3的意义就很小。 因此，它基本上围绕着您对ML模型构建过程的理解-收集数据，清理数据，标准化数据，标记数据(如果需要)，选择模型的能力，最重要的是调整模型的超参数，部署将模型投入生产并确保生产绩效。

So remember, more ML, less AWS.

因此请记住，更多的ML，更少的AWS。

多一点曝光？ (Little more exposure?)

If you have already known the basics, good. Otherwise, let me explain:

如果您已经了解基础知识，那就好。 否则，让我解释一下：

Layman对构建ML模型的步骤的理解？ (Layman’s understanding of steps to build an ML model?)

1. Gather and Get the data
   收集并获取数据
2. Clean the data
   清理数据
3. Label if required, make the data ready for use
   如有需要，贴上标签，准备好使用数据
4. Choose a model/approach/algorithm
   选择一个模型/方法/算法
5. Train the model using the data
   使用数据训练模型
6. Re-iterate through this process until you get sufficient accuracy for the prediction
   重新执行此过程，直到获得足够的预测准确性为止
7. Deploy the model to use it in Production (Enterprise level)
   部署模型以在生产(企业级别)中使用它

那么，在AWS上还有什么可能呢？ (So, what more is possible on AWS?)

All the above steps can be done traditionally on-premise, basically on a powerful local machine or a bunch of servers using a combination of the following tools:

上述所有步骤通常都可以在本地进行，基本上可以在功能强大的本地计算机或服务器上使用以下工具的组合来完成：

Python, Pandas, NumPy, Scikit_learn, Apache Spark MLlib, Google TensorFlow, Keras, PyTorch, Knime, Weka, Jupyter Notebooks, IBM Watson, Orange3

Python，Pandas，NumPy，Scikit_learn，Apache Spark MLlib，Google TensorFlow，Keras，PyTorch，Knime，Weka，Jupyter Notebooks，IBM Watson，Orange3

Similarly, by using the super ability of cloud providers, you can get them done seamlessly on the Cloud, say AWS.

同样，通过使用云提供商的超强能力，您可以在云上无缝地完成它们，AWS说。

On AWS, there are a lot of services provided by AWS in general for various use-cases which helps to accomplish a truly end-to-end ML solution.

在AWS上，AWS通常为各种用例提供​​很多服务，这有助于实现真正的端到端ML解决方案。

Mainly services like S3, Kinesis-Streams, Analytics, Firehose, Kinesis Video Streams, Glue ETL, Crawlers, Data Catalog, Athena, Database Migration Service, Data Pipelines; all these are provided for use-cases on Data Storage and migration which will make it easy for steps #1, #2, #3, #7.

主要服务包括S3，Kinesis-Streams，Analytics，Firehose，Kinesis Video Streams，Glue ETL，Crawlers，Data Catalog，Athena，数据库迁移服务，数据管道； 所有这些都是针对数据存储和迁移的用例提供的，这将使步骤＃1，＃2，＃3，＃7变得容易。

AWS also provides a suite of services with exclusive use-cases in Machine Learning which make it easier for #4, #5, #6. Just a part of them actually.

AWS还在机器学习中提供了一套包含专有用例的服务，这使得＃4，＃5，＃6变得更容易。 实际上只是其中一部分。

AWS made it so brilliant with Amazon SageMaker, which is a cloud machine-learning platform that takes care of abstracting a ton of software development skills necessary to accomplish the task while still being highly effective, flexible, and cost-efficient. To say, mainly it helps you focus on the core ML experiments and supplements the remainder necessary skills with easy abstracted tools. SageMaker supports frameworks like TensorFlow, PyTorch, Apache MXNet, Chainer, Keras, Gluon, Horovod, Scikit-learn, and Deep Graph Library. See the image below for the entire suite of ML Services provided by AWS.

AWS借助Amazon SageMaker取得了如此辉煌的成绩 ， Amazon SageMaker是一个云机器学习平台，可以抽象出完成任务所需的大量软件开发技能，同时仍然保持高效，灵活和经济高效。 可以说，它主要是帮助您专注于核心ML实验，并通过简单的抽象工具补充其余必要的技能。 SageMaker支持TensorFlow，PyTorch，Apache MXNet，Chainer，Keras，Gluon，Horovod，Scikit-learn和Deep Graph Library等框架。 有关AWS提供的整个ML服务套件，请参见下图。

They came up with Amazon SageMaker Studio, the first fully integrated development environment (IDE) for machine learning, which otherwise makes it tedious to set up an end-to-end ML solution. A good point to mention, a new algorithm made by AWS — Random Cut Forest is also one of the algorithms provided on SageMaker.

他们提出了Amazon SageMaker Studio ，这是第一个用于机器学习的完全集成开发环境(IDE)，否则设置一个端到端的ML解决方案就很麻烦 。 值得一提的是，AWS制造的一种新算法-Random Cut Forest也是SageMaker提供的算法之一。

Machine Learning Suite provided by AWS | Source: AWS

AWS提供的机器学习套件| 资料来源：AWS

AWS made it so easy by providing services built centric to highly popular use-cases in the market. Even if you are not aware of how Alexa is built, you can build your own Alexa with a combination of these services.

AWS通过提供针对市场中高度流行的用例构建的服务，使之变得如此简单。 即使您不了解Alexa的构建方式，也可以结合使用这些服务来构建自己的Alexa。

On a high level,

在高层次上

Alexa = Amazon Transcribe + Amazon Lex + Amazon Polly

Alexa = Amazon Transcribe + Amazon Lex + Amazon Polly

AWS上每个上述服务的描述： (Description of each of the above services on AWS:)

Amazon Transcribe — Amazon Transcribe is an automatic speech recognition (ASR) service that makes it easy for developers to add speech-to-text capability to their applications. Using the Amazon Transcribe API, you can analyze audio files stored in Amazon S3 and have the service return a text file of the transcribed speech.

Amazon Transcribe- Amazon Transcribe是一种自动语音识别(ASR)服务，使开发人员可以轻松地向其应用程序添加语音到文本功能。 使用Amazon Transcribe API，您可以分析存储在Amazon S3中的音频文件，并使服务返回转录语音的文本文件。

Amazon Lex — Amazon Lex is a service for building conversational interfaces into any application using voice and text. With Amazon Lex, the same deep learning technologies that power Amazon Alexa are now available to any developer, enabling you to quickly and easily build sophisticated, natural language, conversational bots.

Amazon Lex- Amazon Lex是一项服务，用于使用语音和文本将会话界面构建到任何应用程序中。 借助Amazon Lex，现在任何开发人员都可以使用与Amazon Alexa相同的深度学习技术，从而使您能够快速轻松地构建复杂的自然语言对话机器人。

Amazon Polly — Amazon Polly is a service that turns text into lifelike speech, allowing you to create applications that talk and build entirely new categories of speech-enabled products

Amazon Polly- Amazon Polly是一项将文本转换为栩栩如生的语音的服务，使您可以创建可以对话的应用程序并构建全新的语音支持产品类别

Made sense? Cool.

有道理？ 凉。

一旦您喜欢ML，就相当简单。 (Once you like ML, it is fairly simple.)

All the exam tests us is the ability to express the intuition behind basic ML algorithms.

我们所有考试的目的是能够表达基本ML算法背后的直觉。

Here is a decent illustration of what it takes to build, train, and deploy a model.

这是构建，训练和部署模型所需要花费的一个很好的例证。

Machine Learning Workflow | Source: Google

机器学习工作流程| 资料来源：谷歌

考试不需要什么，但需要学习： (What you don’t need for the exam, but need to learn:)

Nuances can be understood if you know these, otherwise getting your certification done will become just a mug up goal and soon you will give up for the convoluted stuff there.

如果您了解这些细节，就可以理解它们之间的细微差别，否则获得认证将只是一个艰巨的目标，很快您将放弃那里的繁琐事物。

Having a hands-on with Jupyter Notebook, Python Language, some libraries like Pandas, NumPy, Scikit_learn would be beneficial.

动手使用Jupyter Notebook，Python语言和一些库(例如Pandas，NumPy，Scikit_learn)将是有益的。

Knowing TensorFlow or Keras or PyTorch would be awesome.

知TensorFlow或Keras或PyTorch将是真棒。

什么是20小时？ (What are the 20 Hours about?)

Frank Kane and Stephane Maaraek are some brilliant folks in Cloud-based certifications. This is the best and optimal course that you can find in the market.

Frank Kane和Stephane Maaraek是基于云的认证中的杰出人才。 这是您在市场上可以找到的最佳和最佳课程。

This course work is a decent culmination of just required pieces condensed in 10 hours of content. You might take a couple more hours to take notes and also to do some runs of active recall. This is why the title of this Medium story is tailored around hours, once you know the basics of ML, the purpose of ML, applications of ML and basics of AWS, all it takes is a 2x run of this course and finely taken notes and a couple of runs of Active Recall before you take your exam.

该课程的工作是将10小时内容压缩后的所需片段集结起来。 您可能需要花费几个小时做笔记，并进行一些积极的回忆 。 这就是为什么这个中等故事的标题会在几个小时内量身定制的原因，一旦您了解了ML的基础知识，ML的目的，ML的应用程序和AWS的基础知识，所要做的只是本课程的2倍运行，并仔细记录笔记和参加考试前，需要进行几次“主动召回”。

Boom! 20 Hours.

繁荣！ 20小时。

我准备修改的转储： (The dump I prepared to revise:)

I made this cheat sheet to do some runs of active recall to help me persist my learnings, hope this helps. Remember, the sheet makes no sense before you actually learn, but becomes ultimately useful once you finish.

我制作了这份备忘单，进行了一些积极的回忆，以帮助我坚持学习，希望能有所帮助。 请记住，这张纸在您真正学习之前没有任何意义，但是一旦您完成，它最终将变得有用。

Cheat Sheet

备忘单

Good luck to you!

祝你好运！

End.

结束。

翻译自: https://medium.com/swlh/how-i-prepared-for-aws-ml-specialty-in-20-hours-and-cracked-it-ad658bb778bc