First of all, I’m really excited and thankful for such an amazing platform with its tools like fastai library. I’ve been learning ML and DL concepts along with Linear Algebra and Calc as a night reading/explorations after work for a few years. This curiosity raised to the point that I’ve been playing with fastai library recently and getting amazing results.
Yesterday I trained a model with a custom plant disease dataset and got an amazing result and precision (around 0.982 accuracy) and am still in astonishment!
Initially, I started from DL for coders v2 which I suppose uses the latest version of fastai library. Honestly, it seemed a bit chaotic and at times a bit slow for me to follow.
Then I went for DL from foundations and explored there’s a DL for coders v3 which is part 1 of the course.
Part 1 was so well structured for me and I followed along for first 4 lectures which is mainly about vision problems and I really enjoyed them. Also, I got a lot of fresh info and connected a lot of dots in my head
But then noticed it uses fastai older version (0.7?).
Yesterday I I used the latest version of fastai and it seems like concepts around datablocks differ quite a bit. in v3 @jeremy was using ImageDataBunch and in the newer version it seems like DataBunch naming was changed to DataBlock.
Also, I noticed some differences in lr_find, untar_data and download_data didn’t seem to work as I supposed ( this one most likely I was using it wrong)
I admit I’m not fully devoted to the subject and still have a very shallow knowledge on the API, but I thought it would be way quicker to ask a here.
So, here are my questions:
- If I really like the v3 course, how hard to switch to newer version of fastai after?
- Is there any material on changes from 0.7 to whatever the new version is?
- Does the 0.7 version uses TensorFlow (which I suppose v3 course uses)
- Do you think v3 is outdated?
Frankly, it’s a bit confusing as v3 seems older than v2 lessons
I want to thank @jeremy and his team along with the community for making DL
You guys rock!