How I got AWS Machine Learning Certified

Last week I sat down for a challenge and at the end of 170 minutes I walked out with a AWS Machine Learning Specialty Certification. Here’s what I did in case you are curious.

First let’s go through the information made available by AWS regarding the ML Specialty exam. I think it has a pretty good summary of what they are looking for in candidates. You can find basic details such as the duration of the exam and exam fees. There is also the link for scheduling the exam. Under the section titled “Exam Resources” you will find an Exam Guide and Sample Questions.

What to expect on the exam:

1. 65 multiple choice questions.
2. Questions about AWS services
3. Questions about ML processes and algorithms that have nothing to do with AWS

The Sample Questions contain 10 questions following the same pattern and topic distribution as the ML exam. I eagerly tried to answer the questions and verified my answers with the answers provided at the end of the document. By doing so I was able to identify the areas I needed to work on, and I focused my preparation time accordingly. I liked that along with the answers, a short explanation and additional links are provided. Studying these links is not required for the exam.

The exam is intended to test candidates on using AWS platform for Machine Learning, so I broke up the requirements into two sections: AWS & ML.

For the ML part of the exam you need to know:

1. How to frame the business problem as ML solution
2. Common algorithms used
3. The intuition behind the algorithms
4. How to choose the algorithms
5. How to evaluate the results and tune them to get the desired outcome
6. Basically the entire ML process but no need to know the intricate mathematical derivations

For the AWS part of the exam you need to know the following:

1. High level ML APIs provided by AWS
2. Sagemaker
3. Sagemaker built in algorithms
4. How to collect, prepare, process, analyse, and visualize using AWS tools and services
5. How to securely create, deploy, and manage ML solutions on AWS

Since the important aspect is the intuition behind the algorithms, you don’t have go over university level courses. If you are completely new to ML, I would suggest starting with Google developer’s ML crash course. The most popular course on ML is of course Andrew Ng’s Coursera course, although it is not necessary to complete this in order to pass the AWS ML exam, my suggestion is to listen only to the parts of the lectures where he explains the working of the algorithms and the intuition behind the algorithms. If you need deeper understanding of this domain, then the best resource is the book Pattern Recognition and Machine Learning by Christopher M. Bishop (I always have this book opened and nearby :) ).

On the Exam Guide provided by AWS there are some links to resources. Some of it are topics from AWS service called ML(which is being replaced by Sagemaker but somehow still included in the exam), and some reInvent videos. Sagemaker is going to be one of the main focuses of the exam, I strongly suggest that you go over the Sagemaker deveoper guide. You will find questions in the exam on some nitty-gritty details found in the folds of the developer guide. Make sure you know all the built-in algorithms provided by Sagemaker, the differences between them, when exactly to choose them, and try them out! Other topics include the security aspects building a ML solution in AWS. Using the high level APIs provided by Sagemaker. Using services such as Kinesis and Glue for data ingestion and transformation. And all related services. For this, I found it best to follow a course from online learning platforms such as A Cloud Guru or Linuxacademy.

A Cloud Guru’s course for clearing this exam is well organised with quizzes and labs. Just listening to this course will not be enough. You will have to go through all the additional resources provided during each lecture. Experience with using AWS is a definite requirement, if you don’t have that then you can obtain the familiarity of using AWS by doing these labs(and any other labs you can find, because nothing can beat hands-on experience).

Linux Academy is in the process of releasing their course, so I was able to use only the early access portion of the course, which I found to be a good way to do a fast revision the day before the exam.

Make sure you take the practice exam provided by AWS. The exam will be very similar, but I found the questions were much more difficult in the final exam than the practice exam. The AWS Machine Learning exam is designed to fish out beginners, so make sure you are through with the details of the AWS services and the intuition behind ML algorithms. No point in memorizing.

TL;DR

Required to pass the exam:

1. Suggested resources in the AWS Exam Guide
2. Hands on experience with using AWS for building and maintaining ML solutions — use labs to become familiar if you don’t have enough experience
3. Sagemaker developer guide
4. Sagemaker built-in algorithms
5. Intuition behind choosing and tuning algorithms
6. You can expect questions from this section
7. AWS high level ML APIs such as Rekognition, Polly, Lex, etc — knowing what each service does is enough, no in-depth knowledge is required
8. AWS services such as Kinesis, Glue, EMR, Redshift, etc — knowing how to build ML solutions on AWS using these services. This whitepaper is very helpful.
9. The entire ML process —you can use A Cloud Guru’s course and/or Linuxacademy’s course, or any other courses that are geared towards this exam

Not required for the exam but useful:

1. https://developers.google.com/machine-learning/crash-course/
2. https://www.coursera.org/learn/machine-learning
3. Pattern Recognition and Machine Learning
4. AWS classroom training
5. Any other blogs you might come across online

One final word.. During the exam read the questions and answers carefully, pay attention to clues in the question and obvious trap answers. Try answering using techniques like method of elimination. This exam is pretty difficult but not impossible, think of it like a challenging puzzle to solve at the end of which you get a pretty badge like this