Sorry for not being able to share an awesome project like everyone else.
Instead I want to share an interview: I had interviewed @rachel about her DL Journey and fast.ai, I totally forgot to share it with the community. Here is the link to the interview.
Based on the crap to no crap GAN of lesson 7, I tried the same approach to add colors to a crap black & white image. First I downloaded high quality images from EyeEm using this approach.
Training the Generator with simple MSE loss function, I got not exciting results (the only thing it learned is sky has to be BLUE ):
I replied on Kaggle, but wanted to reply here as well: Your methodology very closely followed the intuitions I’ve been working off of, and your code gave me a couple of "ah hah!"s from my own experiments so far. You can see my detailed response there.
@radek I have restarted the kernel.Then also I am getting this error. I fixed this issue by putting NUM_WORKERS =0 and adding padding_mode=‘zeros’
But Now i am getting the below error while fitting the model. Pls suggest how to fix this error.
RuntimeError: Expected object of type torch.cuda.LongTensor but found type torch.cuda.IntTensor for argument #2 ‘target’
I played around with different loss functions for super resolution to see the impact on the generated images. Interesting to see that impacts quality. I’m also trying to figure out how to deploy a super resolution model on Zeit. If anyone knows how to make the app return an image, let me know.
I a brief post on the Naive Bayes classifier (Introduced in the fastai Machine Learning Course, Lesson #10) on Towards Data Science. Hoping some folks might find it useful.
Jeremy, thanks for the suggestion. I read that post on data leakage - interesting; I hadn’t even heard the term before - and it took a while to figure out what he means by “cross validation folds”! I do see where it would be a problem in the scenarios he focuses on (mostly k-fold cross-validation, if I’m reading it correctly) but I didn’t see much there that seems directly applicable to this case, except of course his solution of holding back a validation ds, which I take as gospel and I think is pretty well baked into fastai.
I’ve read a random sample of the dataset article texts (makes for some fascinating reading!) and don’t see anything that might be a marker showing whether it’s ‘real’ or ‘fake’, but maybe I don’t know what to look for? One obvious ‘marker’ might be the words ‘real’ or ‘fake’ in the text, and that definitely occurs, so to eliminate any chance of that causing leakage, I ran with a ‘clean’ df from which I had removed any records where ‘real’ was in the text and the label was REAL, or ‘fake’ was in the text and the label was FAKE. That took the dataset down to about 4800 records (from 6000). It still produced about 98% accuracy.
I previously had run a few more times with the full dataset, and saw a little variability in losses and accuracy, but generally around 98% accuracy, so it seems like removing the ‘real-REAL’ and ‘fake-FAKE’ records didn’t make much difference.
Interesting situation. Maybe fastai is just that good! Although I admit I’m still a little suspicious… Any other factors I should be looking at?
I guess it depends how they were labeled. E.g., if there is a field that was what channel it appeared on, and every time it says “Fox News” the labeler marked it “fake”, then a simply rule looking for the string “Fox News” would get 100% accuracy.
Wow that’s very nice! I’m not sure I can see any impact of attention - but the other 3 all seem to help. Is that what you’re seeing too? Do you see any other benefit of attention (e.g. faster training, higher LR)?
What have you tried for returning an image? You probably want something like this:
For the final update to our aircraft classifier project, I have added the Data Augmentation progressively resized the dataset. We are now at 99.5% accuracy. Also did a little write up on the heatmaps to better understand the lesson.
The following Medium post describes the details of the process.
The accompanying notebook can be found at this gist.