Whats your take on it so far @ecdrid? Also doing the practicals?
@SHAR1 yup my bad saw the earlier videos. I must say the fastai dev notebooks are pretty darn good, Im guessing even the other ones would be good quality.
1 Like
I believe we they(dev NBS) will be used in the second part of the course!
Don’t yet have access to the practicals but have access to that Udacity Course, here, other tutorials, Kaggle kernels…
The lecs are pretty cool, especially the slides …
I must say the way professor explained the Perceptrons and the Backprop combined with Jeremy’s resources for maths, I got a clear understanding and can now myself calculate derivates without any fear!(if the network is of ok size)
it’s pretty good I would say and will watch the remaining ones also after my exams plus the fastai lecs in December (last week)!(exams till 23rd December)
2 Likes
Hi,
Does the Udacity Pytorch course cover creating custom datasets for image/text or they use the inbuilt ones ?
I guess, there is no dedicated section for it.
If you are interested, https://github.com/utkuozbulak/pytorch-custom-dataset-examples will help.
After that if you want to parallelize your datasets, check out https://stanford.edu/~shervine/blog/pytorch-how-to-generate-data-parallel.
3 Likes
@ecdrid are you following 2018 course? they have not yet released videos of each lecture yet.
and how are the assignments? can you comment on that as well?
Hello Amir, the videos are out on youtube and also you can download the slides from their official website. I am taking the course and found it very useful.
The videos are not public. You can find them from the forum pages of each post.
The official PyTorch resources are now highly polished and look very good:
- Cheatsheet: https://pytorch.org/tutorials/beginner/ptcheat.html
- Tutorials: https://pytorch.org/tutorials/
- NLP tutorials: https://pytorch.org/tutorials/beginner/deep_learning_nlp_tutorial.html
- Code: https://github.com/pytorch/examples
10 Likes
After phase 1 of the Udacity PyTorch Scholarship Challenge, I did a compare and contrast exercise by redoing the flowers classification challenge using the fastai library - it’s very similar to the dog breeds notebook from lesson 1.
In this notebook https://gist.github.com/AlisonDavey/5742c87aa45da57511b7b10bb4f8bd51 I can consistently get at least 98% accuracy with just 10 training epochs and only 16 minutes run time, using a P100 on GCP. ClassificationInterpretation is really helpful. It’s good to be back with fastai! 
This researcher has amazing repos for PyTorch and covers a lot of advanced concepts:
The great NLP tutorial covers basic concepts up to BERT.
Also a vision and general DL repo seems to be underway.
1 Like
I’ve been trying to enroll into the stuff but can’t keep my attention focused longer than for a week
I did a compare and contrast exercise by redoing the flowers classification challenge using the fastai library