This repository contains recent notebooks for my work on completing the official TensorFlow certification:
On my first ever visit to the web page at www.TensorFlow.org/certificate, I remember being ecstatic to see that there was finally an official certification for the technology, and I began working on achieving it. I did research and scholarship to verify all of the content and all of the work required to pass the exam, without actually taking the exam and paying the fee, and then did that work, and this repository is the result.
-
[The Exam Guide]: The official TensorFlow Certificate web page: https://www.tensorflow.org/certificate. You start here. Here you'll find the 9-page exam guide. It tells you exactly what will be covered by the exam and also everything that won't be covered, which is crucial. TensorFlow has many uses and a lot of documentation about each use, but a large amount of that won't be on the exam, so identify what that is and then only do that and no more than that.
-
[The Skills Checklist]: After obtaining the exam guide, the most important section is at "Exam details" -> "Skills checklist"; it's on pages 3 through 5, and it has FIVE sections. I would say that the last THREE sections are the most important, followed by section 2 and then one. The first 2 sections are kind of generic about just simply using TensorFlow (debugging, eg.), and the last 3 sections are actually about generic ML concepts (as implemented with/by/in TensorFlow) and you should know and understand all of it: image classification (CNNs), natural language processing, and "time series"/sequences/predictions. All of this is narrowing down the scope of the things/topics/issues you'll need to be concerned with for the exam.
-
[The Guides and Tutorials]: A note at the end of the "Skills checklist" section says, in short, "All knowledge tested in the exam is adequately covered in the guides and tutorials on tensorflow.org". Very important. The web site indeed has specific sections for both guides and tutorials. Reading a certain combination of both will make sure you pass the exam. The guides and tutorials are all just Python notebooks.
-
[The online course GitHubs]: But even despite the included guides and tutorials being sufficient, there is also both a Coursera course and Udacity course that the certificate web page recommends. They both seem to have GitHub repositories at dlaicourse and udacity_intro_to_tensorflow. It's just 2 Git repositories full of Python notebooks.
So it's a good amount of experimentation via the notebook files, along with a fair amount of reading and understanding, provided by the guides and tutorials, and you should be set. You can put all these notebooks in a GitHub repo, and work with all of them via Google Colab. So all of the work that you have to do is exactly 2 things: 1.) a good amount of reading (provided by the guides and tutorials), and 2.) lots of interactive experimentation with the *.ipynb files.
So after reading the guides and tutorials, and going through maybe the Udacity and Coursera courses, and of course satisfying everything else in the exam guide, you can be reasonably confident that after doing these things, you have a good chance of passing the exam. So if you go through the 80 or so Python notebooks in these 4 or so GitHub repositories, you'd be in tip top shape to pass the exam.
Then, you can collect all of those *.ipynb files and put them in your own GitHub repository (just like this one), and then use Colab to work on all of them for free, in the cloud, through the web browser.
Make sure you go back to that Skills Checklist section and thoroughly go over each bullet point. Remember this: that section of the exam guide has mostly bullet points in it: not paragraphs. So make sure to make sure each one is covered, first to last.
-
TensorFlow Without A PhD is a personally recommended GitHub repository I recommend going through to not only prepare for the exam, but is just awesome in and of itself. It's good quality too. There's some good overlap between the exam and this course.
-
Google Colab makes it very easy to do everything you need in TensorFlow, using only the web browser. I have indeed dealt with the various issues regarding just installing TensorFlow, like making sure your CPU has the appropriate instruction set, managing a Python virtual environment via pip, maybe using Bazel to build it from source to build a Python wheel, etc., even Docker. But with Colab, everything is entirely in the cloud and in the web browser, and it's free, and you can
import tensorflow as tfand it just works. It also integrates with GitHub very well, so it's a perfect combination for practicing 100% of everything required for the exam.