Getting the Google Tensorflow Developer Certification

I recently achieved one of my this year’s resolutions and took the Google Tensorflow Developer Certificate Exam. One thing that I found helpful while I was preparing for the exam was blogs of other people’s experience giving the exam and here’s my attempt to fill in the gaps.

There’s a non-disclosure component in the exam’s terms for obvious reasons and probably why there isn’t enough clarity on the exam. I started my journey on the exam webpage which tells you almost everything you need to know about the exam.

What is the Tensorflow Developer Certificate Exam?

The Tensorflow Developer Certification program is conducted by Google and is intended for students, and developers to demonstrate their practical skills in leveraging Tensorflow to build deep learning solutions. The exam is on Tensorflow 2.x and can only be taken using the PyCharm IDE. The exam fee is a 100 USD but Google provides a Tensorflow education stipend you can apply to if you’re a student or unable to afford the exam. However the stipend takes quite a while to be processed so if you need the stipend, you should apply for it well in advance and not when you plan on taking the exam.

What should you know?

The candidate handbook is the place to start which describes in detail everything you’ll be tested on. As the overview suggests, you need to have a decent understanding of neural networks and be able to design and train models in tensorflow for common machine learning problems like regression, image classification, natural language processing problems, time series predictions etc to be able to pass the exam. Unlike what most blogs online would tell you, the exam isn’t all that hard. If you’re not new to machine learning or if you’ve been using tensorflow for a while, you should be able to pass the exam quiet easily. The candidate handbook explains in detail all the topics you’ll need to know and all the areas you will be tested on. The exam gives you 5 hours, at the end of which your saved models will be auto submitted. I was done in a little over two hours. It took me less than an hour for completing most of the exam but the last question took me awhile. The questions are kaggle level datasets and quite easy for someone who’s not new to tensorflow or machine learning in general.

How to prepare?

The most commonly recommended resources on the internet for Tensorflow are the DeepLearning.AI TensorFlow Developer Professional Certificate on coursera, and the book, Hands-On Machine Learning with Scikit-Learn and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems by Aurelion Geron.

If you’re not new to machine learning and only need a quick refresher or an intro to Tensorflow, there’s a much shorter course on Udacity, which covers pretty much all you need.

I had done quite a bit of machine learning on PyTorch a while back and had some experience using Keras before Tensorflow 2.x came along. Since it’d been a while, following everyone’s study schedules online, I too started with the coursera course, led by Laurence Mooney and Andrew Ng. If you choose to refer only one resource, you can’t go wrong with this one as it is a curriculum designed by the very folks who set the exam. I took the whole course which cost around 50 usd, and read the whole Aurelion Geron textbook too. It was only after I’d finished both these resources I came across the free Udacity course listed on the exam webpage and in retrospect, that was all the help I needed. It is free and covers everything listed in the handbook and then some.

If you’re someone already familiar with Tensorflow, I’d say all you need is this Udacity course which shouldn’t take you more than a day. What I found the most difficult was the time series prediction question, (probably due to my lack of experience in the field) and I’d recommend you read the chapter on that in the Geron textbook. Something I read in the chapter really came to my aid during the exam.

If you’re however new to machine learning and Tensorflow, I’d suggest you take your time with the coursera course, do the assignments and work through the labs and that should get you to a good place. If you don’t have the time to read the whole Geron textbook, perhaps try reading chapters 10–15 and if not, read the chapter on time series. You can power through the Udacity course for a quick revision before your exam like I did, if you’d like.

But the exam really isn’t all that difficult and if you aren’t completely new to Tensorflow or deep learning, you should be able to ace it. It’s aimed at students and developers and not experts or researchers. I took a little more than a month preparing for the exam and spent 1–2hrs daily on the prep. In retrospect, it didn’t require all that effort and probably the Udacity course and the time series chapter alone would have got me through it.

How to take the exam?

Once you sign up for the exam, it might take almost a day for the platform to verify your identity and create the exam for you. In my case, it took three hours to get the identity verification mail once I’d filled in the details in the TrueAbility portal. Any government approved ID which shows your name, picture and signature is sufficient. Once the verification is complete, you can buy the exam. My card was charged twice and I ended up with two exams instead of one which took a while to get resolved due to the time zone difference. So I ended up taking the exam a whole day later than I’d originally planned. You have a period of six months to take the exam once you’ve bought it, so it doesn’t hurt to register early.

You’ll need to have the PyCharm IDE setup before you can take the exam. Most blogs recommended that you watch a few tutorials or take some time getting familiar with PyCharm as part of your prep, if you’re new to it. But I found the IDE pretty intuitive and user friendly. It’s pretty much the same as any other IDE, you’ve a code editor, a terminal, a directory tree and a few buttons to compile and execute, all very straight forward. You can download the community edition for free from the Jetbrains website. Once the exam has been created and you hit start, you’ll be given instructions to add the exam plugin to your IDE.

Once the plugin is installed, it’ll restart your IDE onto a new project created for the exam. The exam will have multiple directories, each containing a machine learning problem to be solved using Tensorflow. Each problem has a subdirectory where you need to save the model once it’s trained. The problems are of varying difficulty levels and hence are also scored differently, i.e., the toughest problems having the highest score and so on.

The saved model is evaluated against some predefined validation sets when you click submit and it’ll let you know how many cases your model passed. The ideal is 5/5 and I made sure all my models got 5/5 before ending the exam. You can submit as many times as you want until you’re satisfied with the score.

If you find the IDE difficult to navigate or your machine too slow to train the models, you can simply run your code and train the model on a colab, download the saved model and move it to the right directory for evaluation. I didn’t have any trouble training my models on my macbook air but I did use a colab for an image classification problem because I was afraid it might take too long.

Getting the Google Tensorflow Developer Certificate

Once you’ve successfully submitted all your models and evaluated them, you can end your exam. The result is immediate, even though you can tell if you passed based on if your models passed the evaluations. You’ll get an email announcing the results as soon as you end the exam.

This mail will give you instructions to add yourself to the global tensorflow developers network which shows all the certified tensorflow developers across the world. It’s a place for folks looking to hire tensorflow developers to find you and also just to flex. It’ll take a few more days for you to get your blockchain verified certificate in the mail. I got mine on the third day.

That’s all I’ve got to say on the subject. If you’re here because you’re prepping for the exam, all the best. You know more than you think you do. Ace it!