This is a bit of a high level question. I am pretty OK programmer. Ruby, C++, Node to name a few… python isn’t really the problem for me. I found that it was more numpy, pandas and matplot lib(I feel the most under equipped with, but I can deal) are the libs really making me feel like a junior programmer again(not a bad feel for a small part). Having said this, data science is a new field for me and it’s not my day job…
So I find myself leaning on the library(fast.ai) to much … I then freak out when something breaks and it takes me 3 hours plus to figure out why… I am literally doing one video …every two weeks(10 hours feels like a dream)… and I am struggling to adapt these techniques to similar datasets.
Did anyone who is not a “python & data science full timer” go through something similar? I’m hoping things should start to click more often when I get to videos 5,6,7.
Hey Aditya, thanks for this. I ran through some pandas tutorials playing data frames and fetching, munging data. Your post is helpful, thank you. Definitely still a n00b myself.
How are you finding the course? Are you jumping into the source code for fastai to see what’s going on a lot?
I’m finding good debugging skills are important - I tried to document some of the debugging practices I learned while doing this course here. They are in the first notebook.
No good advice apart from you need to play it by the ear and be flexible to change approach as you learn. I am now going back to square one - rewatching the lectures and will take another close look at the library. But along the way I took a detour into using pytorch directly and writing my own set of tools, etc.
It is a challenge. I am also planning to create little diagrams as I read the fastai code. Talked to a person at work who made extremely fast progress as a programmer and the two things he mentioned was writing things down and using debugging tools.
I think the best advice I can give to someone just starting out with DL and the library - use it just as it is, not trying to mess around with it too much. At least initially. Speaking from experience - I am now planning on going back and filling in the gaps on using the library as is.
Your link is invaluable - thank you. Another good resource and the interactive window in Jupiter Notebook - brilliant.
I must say, I am not sitting in front of my screen moping about how slow my understanding and progress is. I am very frustrated but moving forward still.
I came across this gentlemen’s website and I found it super helpful. There is a ton I haven’t touched but the stuff on handling images was simple and really useful for me. All code examples. Fast.ai does all this for you in easy to use methods. But just understanding how the “cogs turn behind the scenes” makes me feel like I get it.