My Experience with Passing the Google Cloud Professional Machine Learning Engineer Certificate Exam (2023)

My Experience with Passing the Google Cloud Professional Machine Learning Engineer Certificate Exam (1)


I recently passed the Google Cloud Professional Machine Learning Engineer Certificate exam. I figure I’d write a blog post about it. There are already a few blog posts online on this topic. But all of them are one or two years old. In addition, I think in general there is just too little content out there about what it’s really like to take the exam. This blog post aims to share the latest and most realistic experience of preparing for and taking the Google Cloud Professional Machine Learning Engineer Certificate exam.


If you were like me, who stresses about taking exams, the exam logistics is something you want to have well in hand. Knowing what to expect on the logistics details helps you reduce uncertainty for your exam and gives you a peace of mind.

First thing first, you’ll need to register for the exam. You can just google the GCP Professional ML Engineer Certificate and follow the links in the official website. You can take the exam in-person in an assessment center or do it remotely with proper proctoring software setup. I did it in-person. I’d recommend you do the same as it’s quite onerous to set up the remote exam environment and you’ll also have to worry about network/machine failure. The assessment center has technical staff onsite who knows how to handle glitches.

There are usually multiple assessment centers within your region. Select one with high Google Map review. Things to consider: date/time availability, parking, whether they have lockers for you to stow personal belongings, washroom, Covid protocol, etc. Some assessment centers are colleges/universities facilities. Those tend to be well run.

I picked one inside a college in my area. I arrived almost an hour earlier than my scheduled exam time. That assessment center didn’t seem to care about the time slot. After basic ID verification (you’ll get the details about what to bring from the confirmation email of your exam schedule), they just let me in for the exam.

The exam is all done on a computer. There are 60 questions. They’re all single-choice. You have 120 minutes. The software they used has a countdown timer, which is very convenient. Every question is on its own page. You can click “next” or “back” to navigate through the questions. Feel free to go back and forth on questions as your answer to each question is automatically saved. There is also an option for you to mark a question for later. When you click “next” on the last question, the system brings you to an overview page that displays your answers to all the 60 questions. The ones that are marked for later will have a special symbol on them. You can click and go to any question to review and/or change your answer. There is also a “review all” button on each question that takes you directly to that overview page.

Once you’re done, click “submit”. You will see the final result of whether you pass or not instantaneously. You can now call the staff. They will sign you out of the exam and you’re good to go. The official certification link will come in email in about a week. The email also contains instructions for how to redeem the exam-specific SWAG. You can follow the instructions to claim your perks. Here is a picture of mine.

My Experience with Passing the Google Cloud Professional Machine Learning Engineer Certificate Exam (2)

My Background

I should first share a bit more about my background before I talk about the exam itself so that you can calibrate and compare my experience with your own situation. I studied computer science at school and became a software engineer after I graduated. I’d consider myself a generalist. I’ve stepped up to a leadership role for some time but I can still participate in and contribute to coding and design discussions. In another word, my hands-on technical skills may be a bit rusty, but I still (at least I think so) possess the technical astuteness.

I mostly do backend work. I studied ML in school and occasionally train some toy models for fun. I never did any ML for real production use. My team uses Google Cloud, so I’m familiar with its basics, but not so much on its capabilities for ML since my team’s work does not involve ML.

A Few Guiding Principles That Will Help With Your Exam

I’ll share how I prep’d for the exam in just a bit. Before that, I’d like to share a few high level principles that will help with your exam. You will most likely run into questions that you don’t know how to answer. In that case, here is my advice. Think about what Google Cloud wants to achieve with this certificate.

First of all, it’s in Google Cloud’s best interest to not make the exam super hard. They want people to pass. The certificate’s purpose, after all, is to promote Google Cloud. So they’re unlikely to ask very tricky questions. If you find yourself doing hair-splitting or over-indexing on some “clever hints” you find in the question, stop. Take a step back. You may be going too far.

Secondly, Google Cloud wants customers to be successful on their platform. In fact, a lot of the questions start with “You’re a ML engineer in a X company who wants to do Y”. So, really put yourself in the shoes of a Google Cloud ML user. Use your technical intuition to guide your choice. All the good engineering practices apply here: grasp the business problems, start with something simple, iterate fast, build robust pipelines, have good quality control, have good production hygiene, optimize spending, etc.

Thirdly, Google Cloud has a portfolio of services and tools for machine learning. The general philosophy is to use services and tools that are as GCP managed as possible. It’s easier for you since you don’t need to deal with a lot of boiler plates. It’s also good for Google Cloud as the more Google Cloud can do for you the better it’s for their business. For example, use Cloud Natural Language API when you just want to do typical sentimental analysis on common sentences. Use AutoML Image Classification when you have a small dataset with custom labels and you want to up-train from Google Cloud’s image models. Use BigQuery ML to create simple logistic regression when your data is already in BigQuery. Use AI platform Hypertune to run hyperparameter experiments instead of launching and managing multiple training jobs yourself. Finally when you really have proprietary dependencies or special needs like lift-and-shift from on-prem, know that Google Cloud Kubernetes Engineer and Google Cloud VM are general purpose computing platforms that allow you to pretty much do anything.

Lastly, the exam also tests general machine learning knowledge like precision/recall, training/test split, classification vs regression, how to handle missing data, how to handle imbalance dataset, some very basic Tensorflow API, etc. If you feel like this is your weak spot, don’t worry. Remember, the exam is for machine learning practitioners, not for ML researchers. You’ll be fine as long as you have some exposure to those areas. In fact, some of the questions can even be answered with just common sense on statistics.

Prepping for the Exam

I spent about 3 months with 5 hours per week on average.

I recommend first reviewing the official exam guide, and then go straight to trying out the sample questions. They help you come up with an initial assessment of what the gaps are.

A lot of people recommend the coursera course series of Preparing for GCP Machine Learning Engineer Certificate. I did look at that. Personally, I wouldn’t recommend following that series from start to end. Much of the content there is just too rudimentary for someone with basic ML knowledge. What’s worse, many GCP services and tools in that course series are out of scope. So you’ll end up wasting a lot of time.

For the ML knowledge part, I recommend the Hands-on ML with Scikit-Learn and Tensorflow. It has a good mix of theory and practical application. It’s more than enough for what you’ll need for the exam, but it’s a good resource to review nonetheless for ML practitioners. Another source I highly recommend is the Tensorflow official tutorials. They’re super easy to follow. They’ll familiarize you with the Tensorflow API and some of the modern development of deep neural networks.

For the GCP ML services and tools, as an engineer, I think the best way to learn a stack is to roll up your sleeves and try it out yourself. That’s what I did. You can follow my blog post series that contains step-by-step instructions on training, tuning, and operationalizing models, and organizing the various pipeline stages on GCP. It also includes content for managed ML toolings like AutoML and BigQuery ML. The important thing is that you must get hands-on with coding, debugging, deploying, and automating ML on GCP. There is no shortcut!

Finally, to give the coursera courses some credits, I think the MLOps one is relevant and you don’t typically have exposure to that unless you do ML for real production. The production ML system is also interesting, especially the parts that talk about what could go wrong in production ML. That’s also something that you won’t know if you’re not doing production ML.

Last by not least, here are some highlights on the common services and tools on Google Cloud that you should know about. If you’re a Google Cloud user, you’ve probably run into many of them already. They’re all just a google search away so I’m not adding links for them.

  • Automation: Cloud Build, Cloud Composer
  • Data processing: Dataflow in batch/streaming mode, Dataproc
  • Storage choices: Cloud Storage, BigQuery, Cloud BigTable, Cloud SQL
  • Event based services: Cloud Pub/Sub, Cloud Function
  • Services for inference: Cloud Vision API, Cloud Video API, Cloud Translate API, Cloud Natural Language API
  • Services for easy training: AutoML, BigQuery ML
  • Services for custom training: AI Platform Training with prebuilt/custom containers, AI platform Hypertune
  • Services for deployment: AI Platform Endpoint, AI Platform Online/Batch Prediction
  • General compute: Google Kubernetes Engine, Google Compute Engine
  • Important frameworks: Tensorflow, TFDV, Kubeflow


I hope that this is useful. Good luck with your exam. Remember, when you’re stuck in the exam, review the guiding principles I laid out in this blog post. They’ll help make the best-possible choices for the exam questions.

Top Articles
Latest Posts
Article information

Author: Madonna Wisozk

Last Updated: 03/07/2023

Views: 6039

Rating: 4.8 / 5 (68 voted)

Reviews: 83% of readers found this page helpful

Author information

Name: Madonna Wisozk

Birthday: 2001-02-23

Address: 656 Gerhold Summit, Sidneyberg, FL 78179-2512

Phone: +6742282696652

Job: Customer Banking Liaison

Hobby: Flower arranging, Yo-yoing, Tai chi, Rowing, Macrame, Urban exploration, Knife making

Introduction: My name is Madonna Wisozk, I am a attractive, healthy, thoughtful, faithful, open, vivacious, zany person who loves writing and wants to share my knowledge and understanding with you.