Juggernaut
AI
neural network library
SHOW HN2017
54points
19comments
#755of 2,242 in 2017
Top 34%of 2017 by points
How HN took it
RoastedHalf or more of the comments are critical
1 positive, 2 neutral or asking, 0 mixed, 4 critical, out of 7 comments about the project. Thread score -0.21.
Idea: 1 for, 1 againstExecution: 0 for, 2 againstDesign: 0 for, 1 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Appreciation for projects that start small and can be followed
From 1 comment
Keep going! I really like being able to follow projects that start small, as opposed to 'here is my 10000 line toy project. That being said, you will probably get some flak,…
blahman2 on HN
Objections
Too early and raw to be called experimental or shown on HN
From 2 comments
There's nothing to see here. Trivial NN implementations are a dime a dozen, and this one is no different. It's just a partial work in progress; it's not a 'Show HN'; it's just a…
shadowmint on HNCalling this an experimental neutral network is like me calling an opengl demo that renders a few flat shaded polygons an 'experimental 3d engine'. It's not. I'm sorry if that…
shadowmint on HN
Trivial implementation with no real capability, such as MNIST classification
From 2 comments
There's nothing to see here. Trivial NN implementations are a dime a dozen, and this one is no different. It's just a partial work in progress; it's not a 'Show HN'; it's just a…
shadowmint on HNCalling this an experimental neutral network is like me calling an opengl demo that renders a few flat shaded polygons an 'experimental 3d engine'. It's not. I'm sorry if that…
shadowmint on HN
Confusing API: training data inside network and activation fixed per network
From 1 comment
I'm confused by several of the API choices in the example. Why is the training set part of the network? I would have expected it to be a parameter to the train() function. Same…
yorwba on HN
Rust offers little advantage over Python/TensorFlow for neural networks
From 1 comment
I'm a bit confused. How useful is this if: - Rust cannot compile to the GPU - Neural network programs are usually not large and therefore do not need the type safety that Rust…
amelius on HN
Requests
Move training data to the train method
From 2 comments
I'm confused by several of the API choices in the example. Why is the training set part of the network? I would have expected it to be a parameter to the train() function. Same…
yorwba on HNThanks for your suggestions, yorwba. I do see what you mean about the API and I will work on this. For now, if you want to train a model, you need to pass data to the NN struct…
afshinmeh on HN
Make activation a per-layer property
From 1 comment
I'm confused by several of the API choices in the example. Why is the training set part of the network? I would have expected it to be a parameter to the train() function. Same…
yorwba on HN
Classify MNIST before showing it
From 2 comments
There's nothing to see here. Trivial NN implementations are a dime a dozen, and this one is no different. It's just a partial work in progress; it's not a 'Show HN'; it's just a…
shadowmint on HNCalling this an experimental neutral network is like me calling an opengl demo that renders a few flat shaded polygons an 'experimental 3d engine'. It's not. I'm sorry if that…
shadowmint on HN
Questions
What is the goal of the project?
From 1 comment
What is your goal in writing this project?
bmh100 on HN
Is Rust useful for neural networks given no GPU compilation?
From 3 comments
I'm a bit confused. How useful is this if: - Rust cannot compile to the GPU - Neural network programs are usually not large and therefore do not need the type safety that Rust…
amelius on HN> Rust cannot compile to the GPU Rust might not be able to compile to the GPU today, but any language can target the GPU, it is only a matter of tooling support.
pjmlp on HN#1 is not strictly true, although it's not necessarily useful yet: https://github.com/japaric/nvptx#nvptx
kersny on HN
Alternatives named
TensorFlow and Keras
From 2 comments
I'm a bit confused. How useful is this if: - Rust cannot compile to the GPU - Neural network programs are usually not large and therefore do not need the type safety that Rust…
amelius on HNwhich gives you quicker access to the new stuff, but the effort is about the same as writing a new implementation in another language Not really. You can just serialize a…
danieldk on HN
PyTorch
From 1 comment
> All cool neural network research is done on Keras/Tensorflow Add in PyTorch if you are really talking about nn research.
mi_lk on HN
Leaf
From 2 comments
> - Neural network programs are usually not large and therefore do not need the type safety that Rust offers TensorFlow Fold [1] uses static type checking at model compilation…
yorwba on HNAuthor of Leaf here. Yes, Leaf has been dead for more then a year. There recently has been an effort by some community members to pick up development of Leaf under a new name:…
hobofan on HN
TensorFlow Fold
From 1 comment
> - Neural network programs are usually not large and therefore do not need the type safety that Rust offers TensorFlow Fold [1] uses static type checking at model compilation…
yorwba on HN