Reviewing the ‘Tensorflow Developer Certification’ by Google
Everything you need to know about the certification
Why?
Some time ago, I read this article by Andrei Lyskov (Data Scientist at Apple) on Data Science that asked a thought-provoking question —
Should Data Scientists Be Licensed?
While today any experienced data-scientist is known through their public contributions, papers published or through their projects/work, a high number of developers and students have been eagerly learning the science and applying it vigorously to solve challenging ML questions in various domains, such as computer vision and natural language processing.
In his article, Andrei talks about creating a standard through licensing, low-quality workers who cannot meet the new entry requirement would be forced out, while the more driven ones would have to engage in job-related training to meet the new expectations. While this is a compelling yet sensible argument, deciding who is qualified and who is not can be a difficult task.
Data science and Machine Learning has been the fastest area of growth and employment of the 21st century. With more employers looking to hire more ML developers and data scientists, it’s far outstripped by the rising demand in postings, meaning there potentially may not be enough skilled applicants. Putting the problem stated above and the obvious surplus of developers in Data Science and Machine Learning, there is a realization of a potential market in bona fide certifications for developers to claim their place and prove that they are at par with the scholars and the practitioners. Official certifications by organizations have been proven to be influential in other technical areas in Computer Science programming and introducing certifications to the Data Science & Machine Learning developers community may certainly help to assess the question of qualification and maybe a predominant factor in recruitment in the near future.
Pre-requisites
Everything you need to know about the certification can be found here. As promised from my title I will try to review the exam from a developer’s perspective. Additionally, I also took the specialization recommended for this course — Tensorflow in practice. If you are a beginner planning on taking this exam, I would highly recommend taking up the course. Although I work on NLP with Keras, I learned a lot of new concepts in my last specialization (time-series). Andrew Ng and Laurence Moroney (Instructor of the course, Google brain) have perfected the approach and have done an amazing job from start to finish.
The exam review
The exam requires you to set up a Google Developers Certification on PyCharm but currently, for the latest version of PyCharm (20.+) the plugin is not supported (you can follow the issue here). Currently, the only workaround is to downgrade PyCham. I had some trouble setting up PyCharm (since I don’t use PyCharm to develop professionally) and on my first try, the IDE crashed. I was forced to restart the IDE but I could not access my exam due to a possible corrupt or misplaced .idea file.
The Google Tensorflow Certification team inspected the crash and they were generous to help me with a free re-take (in case the same happens to you, please email them — it might take some time but it’s certainly better than failing the exam since it auto submits after 5 hours!)
There are 5 questions on the exam, each one difficult than the previous one. Each question is graded out of 5 marks. Although it is a little unclear on what parameters the questions are graded on (clean codebase, validation/test accuracy, etc) it’s best to experiment with your loss and optimizers to make sure you get the top grade. You will have to submit your answer to move to the question and you will be graded when you submit each solution — however, you can always come back to a previous problem and improve your solution and submit again.
Although 5 hours feels like a lot of time, it is definitely required for test-takers on CPU-only machines as in some problems you will be working on large datasets, and training will take time. So, keep an eye out for that — time management is key.
The verdict
I think this was a much-anticipated certification from Google and the Tensorflow Team. Although the certification covers basic usage of Tensorflow and Keras APIs, I hope that they would introduce advanced Tensorflow certifications with harder problems. Also, I feel the exam follows the Coursera course. If there’s any way it could be a little different from what was taught on the course, it would give the test-takers a better sense of accomplishment. But overall, a great start!
If you have any questions about the course, leave a comment below — I’ll be sure to reply.
Stay safe!
