Are the Jupyter Notebooks sufficient for completion of this course?

I was wondering if there will be some limitations with the NBs leading to rely on other servers (AWS, etc). I read somewhere that as long as the GPU works fine it is ok. In Week 1, I was successfully able to “run” the Cots & Dogs model. Is that enough or I might have troubles later on?
Apologies for seemingly dumb questions :sweat_smile:

In my experience it is just easier to use another server. Colab is a free option. I like Jarvislabs, though that’s not free.

In theory a local GPU is great - in practice it can unexpectedly take a lot of debugging and environment setup time to keep it working. You certainly can use your own GPU, but it will be harder to debug if you have GPU issues.

I am going to use cloud resource so I can focus on the modeling and not futzing with GPU and environment issues.

4 Likes

I’m currently in the process of getting my own DL box up and running again (haven’t tried in years), but I have to agree with Ezno’s sentiments. I still remember the long nights of staying up till 1-something in the morning, trying to get all the various Linux drivers to work, resetting everything, trying again, uttering curse words… While long term I do want to have my own setup, it’s becoming more of a thing I work on as time allows, as it is not a stress-free endeavor.

However, using commercially available servers can be a pain as well. I think there are some parts in the course where you’ll be specifically asked to use Azure, AWS etc. My Azure free trial expired so now I have to go the paid route, and AWS denied my request for the GPU instance I need for this course. For me personally, I’ll be seeing what alternate arrangements I can make for those course segments. Hopefully your experiences will be better than mine. Good luck.

2 Likes

For the piece in the book that mentions azure for the image scraping, there’s another alternative of using duck duck go as well that works well without needing to mess with Azure! Downloading images | Practical Deep Learning for Coders

2 Likes

You should be able to run all the notebooks in Google Colab which is a free option. I have an old setup (Dell T3600 = Xeon 8 core + 64GB + 1070ti) and it is not that much faster than google colab. The amount of trouble I had to go through to set it up, my advice to anyone just starting out would be to look into cloud based options like Google Colab or Paperspace (which I think also has a free tier? at least it used to).

The only advantage I have with my local setup is that I can keep things running for longer than 12 hours. With the free Colab notebooks you have to keep your session alive.

AFAIK/AFAICT , none of the course related notebooks require more GPU than a free colab or paperspace instance would provide out of the box.

My advice to anyone starting out would be to try these things first in colab or paperspace and then look into setting up their own system with a GPU as they are quite reasonable for completing the course. You do not want to wrestle with setting up your own box (unless of course you’re a sucker for punishment like me and like to solve such problems :wink: … )

Good luck!

2 Likes