My Story of Taking the TensorFlow Developer Certification Exam

My Story of Taking the TensorFlow Developer Certification Exam

My overall experience of taking the exam, how I prepared for it, and what I would’ve done differently if I had to take the exam again.

To be honest, I had not known that TensorFlow offers a certification exam until I saw someone tweeted about it on Twitter. I did some research to find out more about the exam and I said to myself: This is going to be my next goal in my journey of exploring the world of machine learning.

In this article, I want to talk about my personal experience of taking the TensorFlow Developer Certification exam and how I prepared for it. I hope that this article will be helpful to anyone of you who are interested in taking the exam in the near future.

Now before I talk about my experience, it makes sense to talk first about what the TensorFlow Developer Certification exam actually is.

What is the TensorFlow Developer Certification?

As you might already know, Google released an open-source software library for machine learning applications called TensorFlow back in 2015. TensorFlow is one of the most popular deep learning library out there right now that enables you to easily build and deploy different kinds of deep learning models at scale.

The TensorFlow Developer Certification exam allows you to showcase your practical skill to build various models to tackle different kinds of machine learning problems using TensorFlow. If you look at the Candidate Handbook of this certification exam, you’ll soon know that you’d be expected to solve problems related to structured data or unstructured data like images and texts.

As this is a practical exam using TensorFlow, then it assumes that you already know the general concepts behind Shallow Neural Networks, Deep Neural Networks, Convolutional Neural Networks, and Sequence Models. During the exam, you’d be expected to implement these machine learning concepts with TensorFlow.

The exam itself cost $100 per trial meaning that if you fail the exam, you need to pay the exam fee again for every trial you make. If you failed at the first attempt, you can retake the exam 14 days after your first one. If you failed at the second attempt, you need to wait for two months before you’re allowed to do your third attempt.

You’ll be given 5 hours to solve different problems within PyCharm environment and soon after you end the exam, you will get a direct notification via E-Mail whether you’ve passed the exam or not. If you’ve passed the exam, your certificate will be sent to you a couple of days later and it will expire after 3 years.

Why Did I Take the Certification

As a person who wants to continuously develop the skill to solve different kinds of machine learning problems, no doubt that TensorFlow is one of the fundamental tools that I often use.

The idea of having a major goal to take the TensorFlow Developer exam really motivates me to continuously develop my skill on how to utilize TensorFlow to solve machine learning problems. Plus, knowing that there is going to be a time constraint while taking the exam really makes it more challenging. The more I know how challenging my goal is going to be, the harder I will learn to prepare for it.

Another advantage is that if you’re trying to break into an AI industry, having a TensorFlow certification certainly will give you an additional credential, although I can’t guarantee that with a certificate alone is enough for you to break into AI. You can share your certificate in your resume, LinkedIn, or GitHub.

Helpful Learning Materials for the Exam

Now you know what the TensorFlow Developer Certification exam is and you probably want to take it in the near future. Next, you’re probably wondering what learning materials that would be helpful to prepare yourself for the exam.

Well, it depends on what you’ve known so far…

If you already know the general concept of Deep Learning, then TensorFlow in Practice Specialization on Coursera is the best way to prepare yourself for the exam.

If you look at the skills checklist in the Candidate Handbook , you’ll find the relevant skills you need to posses to successfully passed the certification exam. It might be intimidating at first to see the skills checklist, however all of those skills are covered in TensorFlow in Practice Specialization.

It costs $49 per month to take TensorFlow in Practice Specialization, but you can actually audit the course for free and you’ll have the access of learning materials as well as the practical code on Google Colab. It is also important to try to recreate the code by yourself from scratch to maximize your learning experience while taking this course.

However, if you’re not yet familiar with the general concept of Deep Learning, then I recommend the following materials to get yourself up to speed before taking TensorFlow in Practice Specialization:

1. Deep Learning Specialization by Andrew Ng
2. Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow by Aur é lien G é ron .
3. CS231n: Convolutional Neural Networks for Visual Recognition by Stanford University .
4. CS224n: Natural Language Processing with Deep Learning by Stanford University .
5. Deep Learning with Python by Fran ç ois Chollet, which the entire book you can read online for free .
6. MIT 6.S191 Introduction to Deep Learning .

Those are my go-to resources and I find it beneficial to learn Deep Learning concept from different resources and point of view as it helps to solidify my knowledge.

The other thing that I find very important for you to learn before taking the exam is TensorFlow datasets. Learn about how data can be loaded with tf.datasets and more importantly, learn how to preprocess data from tf.datasets such that they are ready to be fed into your models. You can learn more about it in the TensorFlow documentation page .

How I Prepared the Exam

In total, I took about three weeks to prepare the exam.

In the first two weeks, I learned all of the materials from TensorFlow in Practice Specialization as well recreated the code from scratch.

Next, as we need to do the exam within PyCharm environment and I’ve never used PyCharm before, then I took my time to learn more about PyCharm one week before the exam. I think it wouldn’t take a lot of time for you to familiarize yourself with PyCharm if you’ve used other Python IDE like Spyder before.

All of the detailed steps to install PyCharm and plugins needed for the exam is explained in this TensorFlow Developer Certification documentation . Now if you want to learn more about the functionality of PyCharm, then I suggest the following materials:

1. Getting Started with PyCharm video series by JetBrains
2. Getting Started with PyCharm documentation by JetBrains

After I familiarized myself with PyCharm and checked that all of the necessary plugins can be installed properly, then I recreated the code from TensorFlow in Practice Specialization within PyCharm environment to make sure that there is no bug when running the code.

My Personal Experience of Taking the Exam

And finally the exam day has arrived, 9th of July 2020…

To take the certification exam, all you need to do is heading up to the TensorFlow Developer Certification website , and click on the ‘Begin Exam’ button. From there you need to provide your ID and payment information, as well as to set up your exam environment.

All of the steps to set up your exam environment are going to be explained in detail in a confidential PDF document once you finished with the payment, so you don’t need to worry about it.

After all of the exam environments are set, now you have an option to begin the exam by clicking the ‘Start Exam’ button. Once you click that button, the 5 hours timer will start right away and you need to finish different tasks within this time period.

Does an access to GPU a must to complete the exam?

Now you probably ask: do you need an access to GPU to be able to complete the exam? My answer would be: not necessarily.

I personally trained all of the models solely on my ASUS laptop’s CPU, so nothing sophisticated there. Until now I don’t know why I didn’t use my free access to GPU on Colab during the exam and I realized it once I finished my exam. But we’re all been there I guess.

But this proves one thing: you can finish and pass the exam although you don’t have GPU. However, if you have a plan to train all of your models on your machine’s CPU, don’t repeat my mistake and I’m gonna tell what my mistakes was.

As you already know from the Candidate Handbook, there are different problems that you need to solve within 5 hours, each problem is going to be more complicated than the previous one.

My mistake was that I solve the problems in order, meaning that I solved the simplest problem first and then move on to the more complicated one after. There is nothing wrong with this approach, but by the time I try to solve the more complicated problem, I don’t have a lot of time left to train the model, although the model is computationally more demanding to train than the previous problem.

I’ve basically done the coding part for all of the problems with about an hour and a half left, however there are two different computationally intense models that are still training at the same time. From there, it is basically the race between the processing power of my laptop’s CPU and the ticking time.

Luckily, all of the trainings were done before the time is up and they perform really well in the validation set, so it’s all good.

Now it leads me to the next point..

What would I’ve done differently if I had to take the exam again

If I want to solve the entire exam entirely with my laptop’s CPU, then I would like to solve the more complicated problems first before moving on to the less complicated one.

Without giving too much detail about the problems, the model that involves convolutional layers as well as LSTM or GRU layers of course will be more demanding computationally than the other. If you solve the more complicated problem first, then you’d have more time to train the model. If the complicated model doesn’t perform as good as you’d expect, you’d still have a time to tweak it and re-trained it. As soon as the model trains, move on to the easier problems.

To avoid the race between the CPU and exam time, I would also like to utilize the free access of GPU using Colab. You can solve the problems using Colab, train the model there using GPU, and copy the trained model to the directory of the exam. Anyone with an access to Google Drive can use Colab for free and if you want more detail on how to setup and use Colab, I recommend you to readthis article.

How Do You Know that You’ve Passed the Exam?

Soon after your exam ends, you’ll get a feedback via E-Mail to let you know whether you’ve passed or not.

However, you should already have an idea whether you’ll pass or not during the exam period. As soon as you submit the trained model in every problem to the grader, you’ll get a feedback right away.

Takeaways

Now that you know all the necessary things about TensorFlow Developer Certification exam, here are some takeaways:

• If you already know about the general concept of Deep Learning, then TensorFlow in Practice Specialization on Coursera is the best learning material to prepare for the exam.
• If you’re not yet familiar with the general concept of Deep Learning, then the learning materials that I’ve listed above will be helpful to get you up to speed before taking TensorFlow in Practice Specialization.
• Learn about TensorFlow datasets and how to preprocess them.
• Take your time to familiarize yourself with PyCharm environment and try to recreate the project from TensorFlow in Practice Specialization in it to make sure that everything is working perfectly.
• If you have a plan to finish the exam entirely on your machine’s CPU, make sure that you manage your time properly by solving the more complicated problems first before the simpler one.
• You can use the free GPU access on Colab to optimize your workflow during the exam.