It looks to me as though some of the code from the EmbUnetModel has been removed. Specifically the class is missing a self.mid_block and the self.ups.
I think that these might have been cut and pasted instead of copy and pasted into the later class CondUNetModel, since they seem to be there and adding them back into the class fixes the problems I was having. Should I make changes and submit a pull request?
Yes please! (And at-mention me in the PR)
This is wierd. I have been working from the web preview of the github and have been seeing the following:
When I forked the repro and opened the file to add the missing code I saw the following:
I have never seen this before. I tried a different browser to make sure it wasnāt just a cache problem but saw the same thing.
Bottom line is that it doesnāt seem a pull request is necessary, just donāt work from the web preview as I was doing (I prefer not to copy the notebooks and run them but to try and recreate then). It seems the web preview isnāt always the same as the underlying code, either that or I am doing something really daft!
I have found the miniai library really useful. Its lightweight, easy to understand, easy to adapt as necessary and very extendable. Post the course I have even found myself using it in preference to fastai on many occasions (for the above reasons). @jeremy Iām sure lots of people have developed callbacks etc to use with miniai but the course repro doesnāt seem to be the best place to gather them together since it is so closely tied to the course. Is it worth creating a separate miniai repro that can be a development area. I know you have lots of repros that you already maintain and this might be one too many, just a thought though.
Might you be interested in creating such a thing? Iād be happy to send people your way!
Hi Jeremy, I would certainly be interested but would need some help and advice along the way. I was thinking that the main thing initially would be to recreate notebooks where each is focussed upon specific things, as in the fastai library. Iāll make a start and see how it goes.
Once the basics are in place I think there are a lot of callback that people on the course have already developed that could then be added
Thanks
Hi John, if you need an extra hand, Iād love to collaborate!
Hi Umer, that sounds great. Iām on holiday this week but intend to make a start on my return so maybe we can discuss the approach to take when I get back
Hi Umer, I am back from holiday. I was planning to use the same nbdev methodology and the first thing I was debating was whether to try and restructure the notebooks into more functional areas or to just follow the existing layout. I would quite like to restructure a bit since the notebooks follow the course and I find its not always obvious where the various classes and methods are. The sort of structure of the full fastai library would make sense to me. Having said that it will increase the amount of work, what do you think? @jeremy, would it be possible to give us a channel for discussion in the forums so that others interested can also join in?
I would suggest just creating a thread here for that purpose
Hey John, Iām currently sick, so will answer in the next few days
Get well soon, I have adopted the approach of some refactoring to make things a bit more suited to the current purpose
Hi John, Iām back! 
I agree we should structure along functional areas. What functional areas have you chosen?
Also, have you created a GitHub repo already? If not, I can quickly do it.
If you want, we can quickly hop on a call to discuss what to do next & by whom.
I have just finished watching the last lesson and will be going through the notebooks from this lesson over the coming week.
That was quite a journey and thank you so much to Jeremy, Tanishq, and John for taking the time to put in all this work to spread the knowledge.
Iām really excited to try to apply some of the stuff I have learned here to embedded electronics (I work in consumer electronics development as a mechatronics+manufacturing engineer). That should be itās own interesting little path, letās see how it goes. I see TinyML and a few other resources that look like a great place to start.
There were a few points in the course where I was very much struggling and considered giving up but decided to carry on. Part of the reason was I made/saw this graph below - it shows how many people saw each particular lesson. I told myself that even if I didnāt do any of the work and merely watched the lessons and understood 30% of it, that would be enough to make me understand things better than a lot of people out there and put me in a better position for my own future (but of course once I watched the lesson video I felt like diving into the notebooks).
Glad I powered through and again a huge shoutout to Jeremy, John, and Tanishq for taking the time to put together something amazing like this.
If the team has any suggestions on what sort of contributions from our side might be helpful to them I would love to know. Iāve already purchased the book as a thank you but not sure if thereās anything else I could be doing.
Hey everyone,
I want to give credit after reaching the final video: you guys are awesome. Thank you for all the videos, for all the effort, while providing everything for free, all the notebooks, all the tips and tricks, plus youāve explained how to write pretty much most of your libraries from scratch without charging a single cent along the way which is super cool if you ask me. Great job! Much appreciated!
To team fastai, Thank you for everything! Itās one thing to teach how to do things using a framework, but itās an another thing to teach how to write a framework from scratch - miniai. I donāt think a comparable course exists out there, even in the paid realm of certification courses.
I am a better programmer and ML Scientist now, thanks to the course.
Hoping to see a series on āNLP - from zero to chatgptā, or something similar
