I watched almost all videos of all courses since 2017, haven’t touched them in the past 18 month.
I want to start again but don’t know were to start from. I still remember the basics, and how everything works but don’t remember any API or code.
I had more success with the Keras version of the course back then than with the FastAI version, for example on the first cat/dog, the Keras code was 12 line of code, but I learned way more than that compare to 3 line of FastAI that did everything internally.
after first 2 video on Keras, I could work on my own projects with different constrains, but I couldn’t achieve the same result with FastAI.
I am wiling to start over and put more effort into it, where should I start from? especially on FastAI library
I think I would definitely go through the most recent fastai v3 course: https://course.fast.ai/
Since you are already familiar with the keras version I think you can skip through the parts that you are already familiar with. Definitely do some notebooks and inspect the code though. If you have stopped early due to already knowing the materials, you will probably learn somethings in the latter videos. Even more so in part 2.
I think for you, the initial concepts that are being introduced are going to be easy, but the actual API of the code is hard. I think you can learn more about the API mostly by getting into the part 1 notebooks, and inspecting the code using “??”. If you have not done a particular project before take the time you are going over part 1 to do a project as well.
On course v3 part 2 of the course, Jeremy pretty much rebuilds the code from scratch. This will be the part that really cements the knowledge you have on the code, and the lower levels of deep learning.
Fastaiv2 is being introduced in coursev4, which has not released yet. This will be a more stable version of the library, and I highly encourage you to take a look coursev4 when it is released.