A really interesting bootcamp on Full Stack Deep Learning took place in UC Berkley late last year and early this year. I believe Jeremy spoke at the Sunday session. They have just released the lecture videos and workshop github repo online.
The materials feel like a super compliment to what I learned on Fast.ai and Deeplearning.ai and I thought it would be useful to share with others on the forum. I have nearly finished the videos and workshops from day 1 and would love to see what people think and possibly start a conversation around some elements of it. (Note: they are due to release the videos for the 2nd day soon).
The topics covered are all more on the practical side of things, including:
Formulating DL problems and projects
Finding, cleaning, labelling data
Picking the right frameworks, tooling and infrastructure
Running multiple experiments in parallel
Troubleshooting training and ensuring reproduce-ability
No - I just saw it from your post. I had been fumbling my way through flask and docker with some success but a lot of frustration, so this is going to be excellent.
I made a little flask api that refers to a react front end to do mnist predictions. I hosted it on render using docker. Check it out here: https://wonderful-archimedes-9efd44.netlify.com/
It was a pretty frustrating process, I’d like to rejig the setup after I’ve done the course.
That is super cool! Tried out a few characters and works perfectly every time
I think it took me about 4-5 days but I did a good few hours each day. I went through the course fully once but my intention is to go back over the material a few times as I convert a few of my jupyter notebooks into proper demo apps like the one you have above. I’m not from a software dev background so I found a lot of the topics they covered super useful for me (e.g. testing, continuous integration, docker etc.). I believe it is their intention to release a follow up course in the near future that goes into some of the more challenging aspects of productionization (e.g. monitoring, live training etc.). It’s always been my experience that there are lots of good courses online that cover the model building/maths but it was great to come across these two courses that are more on the practical engineering side of things.
Thanks man, it gets a little confused with some numbers but its generally quite accurate.
Me too man, I come from a healthcare background and I’m fully self taught so courses like this really help. I’ll get stuck in when I get a free moment,
How did you get the repo working, I couldn’t get it going, just ran into the “ImportError: libcublas.so.9.0: cannot open shared object file: No such file or directory” error, I installed CUDA and cuDNN, still not working
I couldn’t find any information from when the new videos are. They might be from the November 2019 bootcamp. But they also mention a course at the University of Washington from Spring 2020. It looks like the content is similar to the bootcamp from March 2019, but there are more guest lectures now.