Visual introduction to PyTorch
0byte.ioReally nice introduction. Two things stood out to me that I think set this apart from the dozens of "intro to PyTorch" posts out there:
1. The histogram visualization of the different tensor initialization functions is a great idea. I've seen so many beginners confused about rand vs randn vs empty, and seeing the distributions side by side makes the differences immediately obvious. More tutorials should lead with "the best way to understand is to see it."
2. I appreciate that the article is honest about its own results. A lot of intro tutorials quietly pick a dataset where their simple model gets impressive numbers. Here the model gets 18.6% MAPE and only 37% of predictions within 10% — and instead of hand-waving, the author correctly diagnoses the issue: the features don't capture location granularity, and no amount of architecture tuning will fix missing information. That's arguably the most important ML lesson in the whole piece, and it's buried at the end almost as an afterthought. "Great models can't compensate for missing information" is something I wish more practitioners internalized early.
The suggestion to reach for XGBoost/LightGBM for tabular data is also good advice that too many deep learning tutorials omit. Would love to see a follow-up comparing the two approaches on this same dataset.
Thank you so much. Really appreciate the thoughtful feedback!
I've watched many intros. Somehow they always end with 90%+ accuracy and that was just not my experience while learning on datasets I picked myself. I remember spending hours tuning different parameters and not quite understanding why I was getting way worse accuracy. I showed this intentionally, and I'm glad you commented on this!
The XGBoost comparison is a great idea.
Two more recent articles by this author:
https://0byte.io/articles/neuron.html
https://0byte.io/articles/helloml.html
He also publishes to YouTube where he has clear explanations and high production values that deserve more views.
https://www.youtube.com/watch?v=dES5Cen0q-Y (part 2 https://www.youtube.com/watch?v=-HhE-8JChHA) is the video to accompany https://0byte.io/articles/helloml.html
Very nice. Thanks for sharing.
Very nice overview, however just like 30 years ago, neural networks and deep learning stuff is not for me, regardless of the tutorials.
Yet, 2D and 3D graphics feel relatively natural, maybe because at least I can visualize that kind of math.
Are the gradient visualizations not doing it for you?
Of course it kind of breaks down as the gradient can no longer be visualized as an arrow in 2D or 3D space and not all concepts transfer as easily to higher dimensions, as one would hope, but some do.
It is quite different, because one thing is to look to a math expression like SDF and understand the 3D shape that comes out of it, the math behind a demoscene plasma field, or a ray traced shape.
Other is making heads of tails of what a neural network with backpropagation means.
This does an honest good job of walking through the beginnings, I would still say understanding/decomposing a decision tree and going through the details and choices /trade offs one makes with how they prepare the tree like binary split or discrete/binning for continuous data. What reducing entropy means, etc. Maybe even start with parametric versus nonparametric modeling pros/cons. You really get to see how probability and statistics is applied in the formulas that eventually will be thrown into a dot function in python.
There is a lot of content on pytorch, which is great and makes a ton of sense since it's used so heavily, where the industry needs a ton of help/support in is really the fundamentals. Nonetheless, great contribution!
The PyTorch3D section was genuinely useful for me. I've been doing 2D ML work for a while but hadn't explored 3D deep learning — didn't even know PyTorch3D existed until this tutorial.
What worked well was the progressive complexity. Starting with basic mesh rendering before jumping into differentiable rendering made the concepts click. The voxel-to-mesh conversion examples were particularly clear.
If anything, I'd love to see a follow-up covering point cloud handling, since that seems to be a major use case based on the docs I'm now digging through.
Thanks for writing this — triggered a weekend deep-dive I probably wouldn't have started otherwise.
This was quite accessible. If I had to pick one point, I wish there was more "handholding" from gradient to gradient-descent i.e. in the style of the math-focused introduction of the function with one parameter, two parameters etc that was done. It felt a bit of sudden jump from the math to the code. I think the gentle introduction to the math is very valuable here.
Good post. I think you mixed torch.eye with torch.full though
You're right! It's wrongly labelled on the image. Thank you for letting me know. Will fix it.
Interesting article. It would be really useful if you have added a full article title to the page meta data, so it would get bookmarked with title. I assume one does not require GPU to try out simple examples provided?
Hi! Yes, I’ll add all the info very soon. No GPU required. I did everything on a MacBook Air.
Are there other similar tutorials like this going into fundamentals of model architectures for example? Something like https://poloclub.github.io/cnn-explainer/ for example
Thank you, this seems like a very good intro to newcomers! Would be cool if you could continue these series with a few more advanced lessons as well
Thank you! That's the plan. I was thinking of writing a 3D mesh classifier explainer next that'll build on these concepts.
Did not expect to see a 3D model of Quake 2's grenade launcher in a PyTorch tutorial today.
Yes! Love that you recognised it! Way more fun than yet another 3D model of a chair or a teapot.
Very nice, thanks! It’s great to be able to play with viz!
For a deeper tutorial, I highly recommend PyTorch for Deep Learning Professional Certificate on deeplearning.ai — probably one of the best mooc I’ve seen so far
https://www.deeplearning.ai/courses/pytorch-for-deep-learnin...
Good introduction! Building pytorch-lite using python and numpy is the way to go.
Free book: https://zekcrates.quarto.pub/deep-learning-library/
Ml by hand : https://github.com/workofart/ml-by-hand
Micrograd: https://github.com/karpathy/micrograd
Really awesome resource, thanks for posting.
Tiny suggestion: make the visualization for torch.zeros and torch.ones have the same y-axis limits so the difference is visually separated.
Looks like a nice resource for the OMSCS Deep Learning class as well.
Very well done. I learned something.
Cool tutorial :) Any PDF versions?
You can print to PDF from your browser. It ends up looking nice.
Yes, I ended up doing that.
Thank you, Sol Roth