How I learned Tensorflow Developer Certification - the ways to go for it

About me

AI, Python, Tensorflow, Flask Microservices are things that I work with daily.
I love to play with new tech!
And dream to do some good with Machine Learning!

Why I wanted to learn Tensorflow Developer Certification

I love AI, and read too many theoretical books on the issue, but finally decided to put it to hands on!
All theory has to have a lot of hands on practice to really get into your head.
Otherwise - you know it but don't really know it.

How I approached learning Tensorflow Developer Certification

First of all, to get the certification you have to start at https://www.tensorflow.org/certificate.
When reading their Candidate Handbook you will learn a lot about the exam and what is needed to be studied for it.

Then, you HAVE TO do this coursera course :
https://www.coursera.org/professional-certificates/tensorflow-in-practice
It gives you a lot of information about the things you will need for exam.
It also guides you to it gently, finally setting all the theoretical knowledge in your head to hands on expirience.
Though not giving you much practice assigment, I redid all of them in PyCharm on my local machine.
(The exam is in PyCharm)

Then you read "Hands-on Machine Learning with Scikit-Learn, Keras, and TensorFlow : Concepts, Tools, and Techniques to Build Intelligent Systems" by Aurelien Geron of O'Reilly publishing - a great book to understand the ins and outs, though less connected to exam.

Finally, after finishing the course I redid all the notebooks another time - in PyCharm, and built myself some common functions in libraries.

After finishing all of it, I paid my 100$ fee for an exam, and after uploading my id and my photo to verify that I am who I am (it takes around an hour from start to end - they have to approve your id for you to be able to start the exam), I downloaded latest version of community edition of PyCharm, got Tensorflow Certificate plugin.... And went to sleep.
Exam is supposed to take up to 5 hours! do it when you feel refreshed. Not when you tired.
So, next day, after coming back from work, I told myself that it's 5PM, and till 10PM I will be in top condition. So I started the exam.
Exam is 5 questions that you have to complete tp real model. Work and rework on those notebooks finally paid back, and after 2 hours I had finished 4 models of 5, with 5th running.
Finally, with last model had some forth and back, and after playing for another hour and half, I understood that I reached top score in one of my tries, restored that point - always save your models in between - and finished exam.
In like a minute time I was notified that I succeeded with it, and now I am official 3 years owner of certificate ( you have to recertificate once in 3 years...) !
All in all it took month and a week, and 149$ (49$ a month for Coursera Specialization - finished it in a mont, 100$ for exam itself).
Important part for people who never worked with PyCharm - play with it! before you start the exam. It is somewhat different than writing in notebooks, and code is running differently.

Challenges I faced

1. Installing Nvidia CUDA drivers for Tensorflow. Took some time, and a lot of games.
2. On the day of exam discovering that CUDA drivers installed where good for Tensorflow - latest version 2.3.0 - that I played with beforehand. The exam is in Tensorflow 2.0.0. And the models saved in 2.3.0 weren't backwards compatible!!! so, in the end after spending 30 minutes of exam time on trying to set up things, I told myself screw it, and run it all on CPU.... Took it's sweet time.
3. Training model for a long time only to discover that it's not standing in how model should be when submitting it to server for check. Lesson learned - run model for 2 epochs, then send it up.
4. Keep in-between models. Especially if you got something like 4/5, so any further attempts won't make you retrain all from start to get that 4/5.
5. Took me time to find how to "Submit and Test Model".... It was inside the assistant, and assistant opens for "starter.py" in relevant directory. Only from there I was able to see my points after submitting model - 3/5, 4/5, 5/5... Even 1/5 is good for trying to run a model before you play with hyperparams.

Key takeaways

Would do it again.
Thought myself too old to go back to getting certificates, only to be surprised that I am able to learn - thought differently - from viewpoint of my expirience.

Tips and advice

1. Know your PyCharm
2. Working locally - install tensorflow and all other entries with exact version as in exam. Install CUDA for them. Check that you are running with GPU - if you have CUDA compatible graphic driver.
3. Try to play time and again with models, with different approaches and prepare libraries for yourself beforehand.
4. During exam use all the internet resources that are accessible to you - this is how we work DAILY, and it's not cheating.

Final thoughts and next steps

Do it.
Fix the knowledge from the theory to practice!
I did it, and so can you.