I'm driving crazy

Continuing the discussion from Wiki: Lesson 2:

Totally agree with the review of this guy. It’s totally crazy the way you have ordered the course and the materials. I thing I’m going to give up because there’s not a right way to go straightforward to every issue.

I’m sorry, but my time’s worth much more then all this maze.

Since you linked a post of mine, I do infer that I’m the “guy”. :slight_smile:

I think you misunderstood. Mine was just a question, not a review.
I’m utterly satisfied with the course, which, don’t forget, is offered for free. It’s also cutting-edge, and the folks behind it are led by the former head of kaggle.

That said, the video and the nb are essentially the same, it’s just the dataset that changes. I wrote down the nb for dogbreeds challenge, let me know if you need it.

My advice is not to give up.

Nothing is easy in this field …
But to give a aroma,

Jeremy taught us RF completely and now I have - turned my understanding into other Boosting tress available…

  • The docs made more sense to me now than previously (I always just used the signature without thinking for a second what am I doing)
  • Understand Metrics better…
  • Increased creativity
  • Searching for the concerned Research paper associated…
  • Learnt how to Overcome the fear of Coding

So yes This course is really for Crazy People as it needs devotion…

5 Likes

It’s not super difficult in terms of coding (you are given working code supposedly), but when things don’t work as expected it can get frustrating. Not having a lesson2.ipynb is rather misleading, and lesson2-image_models.ipynb takes hours to run just for a working example.

First … STOP driving :slight_smile:

These concepts are going to be foreign and quite difficult to even grasp initially. I’ve been doing this for a while and feel lost at times myself. That’s when I buckle down, watch the videos again, get on the forums, google for more insight, struggle, try utilizing what I’m beginning to understand in kaggle comps … rinse and repeat.

How easy it is going to be from one person to the next wrt the course is going to vary for every individual. It could be, that you should focus your efforts on some of the basics (e.g., python, pandas, numpy, using sckit-learn) since they are so fundamental to everything … or it could be, that you may need to watch the videos 2 or 3 times.

Either way, you’re going to feel lost at times learning ML and DL … and I can’t think of a better place to be lost than the fast.ai forums.

Good luck in whichever direction you go.

4 Likes

Like I said, I can give you “lesson2.ipynb”, but the text parts are written down in Italian. I didn’t foresee other people using my nbs.

1 Like

Even a stub (with some time estimates) would be helpful but right now I am dealing with the unexpected issue of the image_models example working but taking way too long to complete. I am guessing it is not using the GPU properly after I updated the graphics card driver without rebooting the system, so I will restart and see if it gets better. Lesson 1 is still working, so no clue what is going on.

You should NOT update the graphics driver. You should use the one which is distributed along with cuda installation.

Let me get it tidy, then I’ll attach it to a PM.

I updated it for other reasons (games). I did not install Cuda at the operating system level, so anything related to cuda must have been installed as part of the conda fastai env setup.