AWS Machine Learning Specialty: you can do it!
At the end of November, I got AWS Machine Learning certified. In this article, I would be happy to share my personal experience. I hope my tips will help you to prepare for the exam and after reading this post you will know what to expect.
Who should take this exam?
As written on the exam’s official page, candidates are recommended to have “At least two years of hands-on experience developing, architecting, and running ML or deep learning workloads in the AWS Cloud”. In my opinion, it’s not a strict rule: consider focusing on practical skills than on years of experience. But you should be rather confident with AWS SageMaker and some other common AWS services like AWS S3 or AWS Lambda.
Keep in mind that this exam corresponds to an expert-level deep dive into Machine Learning AWS services. So you definitely need a well understanding of the logic behind common ML concepts, algorithms, and best practices. You should be able to perform hyperparameter optimization and have hands-on experience with ML and deep learning frameworks.
Is it worth it?
Although this exam is a tough one, the benefits are very cool! Cloud services are very hot right now, and getting officially certified by one of the top cloud providers is awesome! This proves your ability to work with AWS ML-related services and can considerably boost your career.
From the AWS side, after passing an exam you will receive a free practice exam voucher as well as a 50% discount exam voucher to apply toward recertification or any other exam you plan to pursue. You will also get access to the big and growing community of AWS Certified professionals on Linkedin. And you will be able to apply to become a Subject Matter Expert to help decide exam topics, develop questions, and determine passing scores. Almost forgot, you will get access to exclusive AWS Certified merchandise as well.
Preparation best practices
Depending on your schedule, the preparation could take from one to several months. Consider it as a marathon: allocate time every day to learn and practice, don’t miss it! Try not to postpone the exam date: in such a case you will quickly forget many details and be forced to learn them again.
Personally, I found very motivating the 66daysofdata challenge launched by Ken Jee. It helped me to stay focused and find time for preparing every single day for 66 days. I shared my path on Linkedin posts (here is my page), and many people reacted and commented on them. This gave me a sense of community and motivated me to continue my work.
I strongly recommend passing a specified course on Udemy. Its creators framed their knowledge to a well-structured and shareable system, you can consider this course as a starting point for your preparation. I also found useful this official guide, where all exam-related topics and services are mentioned. You can use it as a checklist. AWS also has a nice free Exam Readiness: AWS Certified Machine Learning — Specialty course.
There are 4 domains on the exam: Data Engineering, Exploratory Data Analysis, Modeling, Machine Learning Implementation and Operations. If you want to have an idea of what kind of questions to expect, you can refer to these official sample questions, or even take an official AWS Practice test which is free now.
It is very important to have hands-on practice! Whatever you study: data engineering with AWS Glue or XGBoost hyperparameters tuning on AWS SageMaker, try to make them work on AWS.
But don’t forget to delete/stop resources after using them, otherwise, it could cost a lot! If you want to save money, there are labs on A Cloud Guru, you can pay only for A Cloud Guru subscription (or even start with a 7-day free trial) and don’t pay for using AWS resources.
The last, but not least preparation resource — examtopics. (If you can’t access questions, try a VPN). People say they had on the exam 60–70% of questions from this website. In my case, I had 50–60% of questions from there. However, pay attention that the answers provided are not always correct. Check all discussions and browse for the right answers.
Try to understand the logic behind each question. Pay attention to details — they determine the answer! Catch common conceptions, for example, using AWS Kinesis Analytics for a simple data transformation task is more preferable to running an EMR cluster with Spark.
My personal tip is to make screenshots of questions/answers and to store them in a special folder. Hard-to-remember information such as precision/recall/specificity can also be stored there. You will be able to do a quick review a few hours before the test.
Ready? Schedule your exam!
As was already said, try to pass the exam when your knowledge is fresh. AWS allows scheduling your exam even for the next day or days off.
Go to the AWS Certification portal and sign in (you need your Amazon account, not AWS). Pay attention to the My Profile section: your first name and last name in your AWS Certification Account must match the identification (ID) you will present at your exam appointment.
If you are a non-native English speaker, you can request additional 30 minutes for your exam (before scheduling one). Click on one of the indicated options (marked red in the picture) and select “ESL +30 MINUTES” from a drop-down list.
Then you can register for an exam. Click on the corresponding button, find AWS Certified Machine Learning — Specialty, and choose between “Schedule with PSI” and “Schedule with Pearson VUE” options. I used the Pearson VUE option because I was already familiar with it. You can refer to these videos in order to know what to expect during your exam.
Follow all steps: choose language, read and agree with official policies, choose proctor’s language, select date and time, and proceed to checkout.
Once you have paid for your exam — congratulations! You will receive a couple of emails with payment confirmation and check-in information. In case of any force-major circumstances, you can cancel or reschedule your exam at least a minimum of 24 hours prior to your appointment.
Important rules to consider
It’s important to plan ahead and take care of the following technical components: your computer condition, quiet testing space, official ID, and timely check-in.
Run a system test a day or a couple prior to the appointment. You don’t want to discover that your camera doesn’t work right before the exam! Make sure no one will enter the room / speak loudly in the shared room during exam time. Clean your desk (and room — you will send several photos of it before the exam starts).
Prepare your ID. Make sure your phone is not in arm’s reach during the exam. But keep it relatively close if the proctor needs to contact you. You can start your check-in 30 minutes before the appointment time.
Food and drinks are not allowed. When you take your exam, don’t cover your mouth and don’t speak. You are not allowed to stand up or quit the room during the exam. I wasn’t allowed even to look away (I did this automatically when I was thinking). These rules are rather strict, but, otherwise, the proctor can’t be sure that you are not cheating.
What questions I had on the exam
I had 65 questions and 220 minutes for answering them (I requested additional 30 minutes). It took me 2.5 hours to complete the test. Most of the questions were designed to have one correct response, but multiple-choice questions occurred as well.
Personally, I found the ‘Flag for review’ option very useful. It gave me an opportunity not to be stuck and to refresh from hard thinking. Some of the answers became obvious during the second look.
To my mind, the solid knowledge of AWS SageMaker is checked in the first place. You need to know technical details like file/pipe mode input, types of instances (CPU/GPU, their differences, etc.) Also, it is critical to understand how does AWS SageMaker work with Docker containers under the hood and how do security requirements can be fulfilled.
AWS Kinesis, Glue, Lambda, Athena are also important, but on a less detailed level. AWS Forecast, IoT Greengrass, Amazon Ground Truth, and other high-level services like Lex are less common, with only 1–2 questions for each.
To pass the exam, you also need to understand how to group together different AWS services and make end-to-end business solutions.
Based on the questions types that I had I can advise reviewing the following topics:
- ML algorithms types and corresponding business tasks (you need to understand well where to apply regression/classification/clustering)
- Key metrics to evaluate algorithm performance (accuracy/precision/recall/etc. — how to calculate them and when to use)
- Over- and underfitting problem and how to solve it
- Deep Learning basic concepts, layers types, activation functions, learning rate, batch size
- AWS SageMaker algorithms types (and where each of them can be applied)
If you are not confident
When I saw examples of exam questions after the preparation course, I was upset and thought that it was probably a bad idea to pass it in a week. But I decided to study all the questions carefully and understand their logic. Approaching the end of the questions list, I realized that I was giving correct answers rather often!
I got 873 out of 1000 points. The minimum passing score is 750. You won’t need to give correct answers to all questions, so don’t worry if you don’t know something during your exam. Try to eliminate definitely wrong options and make the best guess. Wrong answers are not penalized.
You will know whether the exam is passed right after the appointment. A few days later you will receive an email with your score as well as your certificate. Congratulations! You rock! Drop a comment if you have already passed any AWS exam — was it hard? If you are still preparing, what topics are the most challenging?
