I can’t on Colab, what’s your setup like? torch, torchvision, etc. Are they all the latest version?
Had the same, seems that calling get_image_files(path) gets images from the images folder and for some reason it pulls images from the annotations folder. Changing your line to
path, get_image_files(path/'images'), valid_pct=0.2, seed=42
should solve it.
3 Likes
I’ll join @mgloria and @muellerzr - similar status of fever for several days and on strict home quarantine.
4 Likes
Oh no! I hope you better soon 
try using pip install ipywidgets --upgrade and check the ipywidget version with:
import ipywidgets
ipywidgets.__version__
the ipywidget version should be 7.5, because they introduced the FileUpload() with that version,
if that doesnt work, try:
pip install notebook --upgrade
and restart your kernel
1 Like
Thank you! That worked for me 
1 Like
Scary to see the exponential increase. I just saw another graph, as expected, also showing the exponential increase, broken out by country. Then I had a closer look: wait a minute, this graph shows the “daily” new cases, not the “total” new cases, how can it still look exponential? Then I realized: of course, the derivative of an exponential function is also exponential!
3 Likes
Dan, if you are so inclined you may like Dr Andrew Ng’s Deep Learning courses on Coursera. I feel that watching the specialization lectures helped me grasp some of the “theory” you mention. Revisiting Jeremy’s lectures helped me appreciate Jeremy’s approach even more because the top down approach really does work and I could appreciate the explanations so much more after knowing some basics of forward/back propagation, activation functions and whatnot.
IMO, there’s definitely theory behind deep learning (and lots of it). Are there unknowns and mysteries? of course, this is a field in its early stages, so, plenty of mysteries still, yet basic mechanisms that inform the current approaches seem to be well understood (by those who do, not mere mortal hobbyists like me.)
This graph gives me hope about the Canadian response, I hope their measures do flatten the curve we need it.
1 Like
I was wondering if anyone wants to join hands and work with me to create DataBlocks for Kaggle top-10 or top-50 competitions? Would be a great way to practice and also help the community in general.
12 Likes
I am doing something similar to it. But I am not stopping with data blocks, I am trying to create a solution itself, probably that could place in top 1 to 5 %. It is actually important to create a complete pipeline, to understand if the data block pipeline is working correctly.
2 Likes
Are you doing this for most competitions?
1 Like
Started with TGS which I participated few years back. My idea is to pick few from different domains from kaggle. I have few competitions in my mind, but if you have already started to work on something, I can join.
1 Like
Excellent, that’s my idea too! To work and pick examples from a few different domains. But I am learning DataBlocks API as I go too. Should we do this on Zoom? Let’s get on a call? And we can work together as a Study Group or maybe Discord?
1 Like
I have started with BengaliAI competition.
Wow thats cool, I was thinking to avoid kernel only competitions inorder to avoid the GPU restrictions that kaggle has.
I have free GPU available from work. So we can share resources and I can share my screen. Needless to say if we do end up writing a blog, due credit will be shared
1 Like
Personally I think this will be a great project because there is no central place for DataBlocks API in Fastai2 with multiple examples at the moment.
Not sure if we should move it to the non-beginner part. 
may be in the Mid-Level API discussion.
I agree with @VishnuSubramanian but I’m not building a top 5 solution. I’m just trying to build a full pipeline. Working with https://www.kaggle.com/c/plant-pathology-2020-fgvc7/data
2 Likes