I wrote the AWS Machine Learning Specialty Exam last week and passed. I am just 1 year into the AWS Cloud and this is my first ever AWS exam. So if you are new to AWS and considering taking this certification then this article will give you some tips on how to go about preparing for the exam.
Remember as a Specialty exam, it is not an easy exam to pass, as it tests both your Machine Learning and AWS knowledge. And it is for this same reason (this blend of skills), that I was attracted to taking the certification. Therefore I could not wait to just go straight for it without going through the normally recommended exam path of first sitting for the Cloud Practitioner Exam, next, the Solutions Architect Exam, etc before Specialty Exams. One of the best ways to show your mastery of machine learning and at the same time position yourself in the modern era of cloud technologies is to take this Exam, offered by AWS.
So I am going to share my experience. How I prepared for the exam and at the same time try to generalize the model to anyone who is not coming from the same context as me. Because we all have different contexts, and so what works for me may not work for you 100%.But getting that “Passed” phrase after 03 hours of taking the exams is what unites us all.
So what was my specific context before the Exam?
Here are 04 things about me that define the context in which I was preparing for the exam:
- I had little knowledge about the AWS cloud. So I knew I had to spend more energy learning AWS.
- Never sat for any AWS exam before this one. So I knew it would be an uphill battle since everyone around me was against me going straight to the Specialty Exam without passing through the Cloud Practioner and Solutions Architect Exams.
- Was already comfortable with machine learning before. This was my main source of confidence
- And finally, what started as just preparing to crack exams, I fell in love with the process and with Sagemaker, especially since I was sharing the knowledge with my subscribers on social media, on my youtube channel, and in meetups.
Alright…Enough about my context, at the end of the day we all want a “Passed” phrase after 03 hours (…while some want to have a score of 990/1000….). So let us look at what the exam is testing and what you should expect.
What are the pre-requisites for the Exam?
First of all, let us discuss what the pre-requisites are, and it is best to copy what AWS says about the pre-requisite and paste them below:
AWS Certified Machine Learning – Specialty is intended for individuals who perform a development or data science role and have more than one year of experience developing, architecting, or running machine learning/deep learning workloads in the AWS Cloud. Before you take this exam, we recommend you have:
At least two years of hands-on experience developing, architecting, and running ML or deep learning workloads in the AWS Cloud
Ability to express the intuition behind basic ML algorithms
Experience performing basic hyperparameter optimization
Experience with ML and deep learning frameworks
Ability to follow model training, deployment, and operational best practices
So now that you are aware of the recommended pre-requisites, let us look at the structure of the exam.
What is the structure of the Exam?
What is the structure of the AWS Machine Learning Exam?
Below are the different domains tested and the weights they carry in the Exam.
As you can see in the table above, there are 04 main domains. You can find the details below:
i.) Data Engineering (20%):
Here the focus is on bringing (ingesting) data into AWS from multiple sources, transforming the data, and storing it.
So knowledge of services like Glue, Kinesis, S3, and Spark will be tested. Remember data is the key asset and without it then there is no machine learning.
ii.) Exploratory Data Analysis (24%)
Here the focus is on data cleaning (preparation), feature engineering, feature selection, data normalization or standardization, visualization, etc.
In fact, this is where AWS skills will not help you. Data analytics is what will save you here. Your ability to clean and find patterns in data.
iii.) Modelling (36%)
The biggest part of the exam is here. Which model will you use in a specific scenario? How will you perform hyper-parameter tuning? How will you measure performance? Etc.
To get it right, you need to have some hands-on experience on some machine learning projects, especially using popular algorithms like XGBoost.
iv.) Machine Learning Implementations and Operations (20%)
Here you are focused on choosing the right resources, permissions, and settings to be able to benefit from running your workloads in the cloud. Your model has been deployed and you need to monitor its performance against other variants of the model and against data drift, etc.
So because your model is running you need to consider scalability, security, fault tolerance, etc
So using your AWS knowledge from Solutions Architect is helpful here. Otherwise learning to apply these settings to your model is what is needed. Bring your model to live strategies to deploy the model in production and a bunch of security best practices.
Enough of the contents… so how do we prepare for the exams?
So how did I prepare for the Exam?
In my specific case I fell in love with the process of preparing for the exams, so I probably did more than was necessary just to write the exams. For example, for almost all the services I came across while studying, I practically deep-dived into all of them, by watching tutorial videos or reading technical documentation about these services. Also practicing them through my AWS account.
Typically, we all do not need all that depth, to succeed, but below are the most efficient things to do which will help you pass the Exam.
This Udemy course, taught by Frank Kane and Stephane Maareck is your number 1 resource when preparing for this exam. You will get lots of tips from Frank as you study this course which will help you pick the right answers during the exam. Normally this will not teach you everything for the exam, since no single course can, but it will teach you the most about the exam than many of the resources combined.
This is also very important as you are touching base with AWS itself to see what they want you to know. Some services are well explained while others are suggested for further reading. Even though I followed through with all the further reading as suggested, you would not always need to. Where you feel uncomfortable just read more about that aspect.
Also, you would get a few sample questions, which look like what you would see in the real exam.
The biggest thing about them is the labs. So if you can afford it, subscribe and complete the labs. This will give you a feel for the Machine Learning services and how they are best used on AWS to solve business problems.
- AWS Reinvent videos :
If you do not have the possibility to take A-Cloud Guru practice labs, reinvent videos have a lot of case studies you can follow along, so you see how those services are used to solve real problems. Here you would also see best practices in Machine Learning on AWS which the exams would be testing you for.
After studying and building all your notes from the previous resources, it is time to start getting your hands dirty by completing many exam-style pass questions.
The best thing about this platform is to use the review mode to do your practice. That way after each question you get instant feedback about your answer and even links to AWS documentation for further reading. This helps a lot to clear any confusion you have about why your answer was wrong. From the corrections and links to documentation, enrich your notes with some clarification of your doubts, such as firehose can’t convert CSV to parquet, whereas Glue can. You may have overlooked this in your notes, especially if you did not do enough labs.
Finally, you will get your score per section at the end of the practice exam, and you can use this feedback to know which sections to focus on to improve your overall score.
I advise this to be your last. These are exam-style questions that have been dumbed here and many of them have wrong or conflicting answers provided by the public as it is open source. So because you have come this far, you will most likely have enough knowledge and conviction to identify the wrong answers that the majority has voted for. And when you visit the discussions around each answer, some have supporting documentation to convince you about the choice of their answers.And also help reinforce your knowledge. Do not forget to take notes for every new findings.
Above are the core things to do and you would stand a high chance of cracking the exams.
And finally, even though these practice questions had a few questions to help me garner a few points, if I had to do it again, I would probably skip purchasing them. Just for the fact that they were not very close to the kind of questions, I saw in the exams. They might need to update their questions to get closer to the way questions are set in the real exams.The exams have many lengthy questions and the kind of context that is being tested is in these practice questions below not very close to the questions in these tests.
Anyway, they helped me grab a few points, though, but if I was only going for a “Pass” and tight on budget, I could manage to bypass them.
Now let’s go to the exam itself.
So on the Exam Day?
The advice of trying to have a good night’s sleep is still very valid, so you stay sharp during the exams. As for me, I even did some exercises in the morning of the exams to pump up and energize myself, which I found beneficial. But you can skip the exercises.
Before the exams, ensure you leave enough time to revise all the notes you have been taking since the beginning. This is the time you need to take one last look at them, so they stay fresh in your mind as you get into the exams. You might remind yourself that Amazon Rekognition could also handle topics and classification, for example. Things you can easily miss if you have not been using Rekognition and might not stick when you read them for the first time.
The exam contains 65 questions, to be answered within 180 minutes. When writing the exams, my advice is to go quickly through all the questions, while flagging the confusing or difficult ones for review. Remember some questions are what we called “Examiners Questions” which just test questions against future exams and will not count in your current exam score.
With that understanding, you need to attempt all questions and do not stress if a question looks out of scope because it probably is one of those “Examiners questions”.
Review all your answers at least two times, before ending the exam. There are some answers you selected when going through the first time, whereas upon critical review while you are reviewing all your answers, you might need to change them.
If you just want the minimum then focus on the AWS Machine Learning on Udemy, AWS Readiness, and then practice with Tutorials Dojo and Exam topics. Try to do as many labs as you can to get a feel for the services. Always update your notes with new learnings or things you thought you knew, but had forgotten or could not recall them quickly as the answer to a question.
Then on the day of the exam, make sure you are well-rested, go quickly through all the questions within about 80 mins maximum, then use the rest of the time to review all the questions from the beginning, while critically thinking through your initial answers to the questions. And if you still have time, you can go through the flagged questions for the last time and make sure you answer them with your best guess. And finally, you can end the exams.
Hope to see you on the other side soon.
Wish you Good Data Luck!!!