Source for simple_neural_network.py is member-only

[tianyizheng02]

What would you like to share?

I want to try to improve the implementation neural_network /simple_neural_network.py because it's currently one of the slowest files by far during build tests and because I find some details of the implementation to be sorta strange. However, the source referenced in the file is a member-only Medium article, so I can't check how the algorithm was originally derived and implemented. If someone here is a Medium member and can access the original article, please relay its contents to me so I can verify the implementation. Going forward, we should definitely find a similar, free-to-read article to replace the currently referenced source.

Additional information

No response

[tianyizheng02]

Ok, I was finally able to take a look at the source article because Medium lets you view a limited number of member-only stories per month. The implementation in simple_neural_network.py doesn't follow the article at all and is far more simplified, which the algorithm file doesn't make clear. Furthermore, the article only discusses forward propagation, and there's no source cited to explain backpropagation as the actual "learning" in the NN. In hindsight, some variable/function names in simple_neural_network.py are outright incorrect and some implementation details are pretty questionable.

[anchor54]

hey sorry for asking this, but since it wasn't clear from the description or your comment, what is it that we are trying to achieve in this issue?

[tianyizheng02]

At this point, the goal is to find a new article that explains neural networks clearly and to rework the existing implementation in simple_neural_network.py based on the contents/explanation of that new article. The current source article is member-only, and the current implementation has some issues.