Hello World of neural nets

Of course, the first self made neural network must be the one that recognizes handwritten digits.

Armed with Neural Networks and Deep Learning textbook and The Rust Programming Language book I slowly converted the code from the first chapter from Python to Rust.

Got little understanding of sigmoid, structure of network layers, gradient descent and ndarray crate.

The shown efficiency is 96.5% against mnist dataset but in practice it doesn't recognize shit is not reliable.

Main lesson learned is that input data format must match the training input as close as possible. Had to add preprocessing to canvas pixel data to simulate mnist inputs to get it to work (sort of).

Hopefully I can decrease error rate with the help of knowledge from the remaining three chapters.

The thing can be challenged here.

Birds recognizer can be challenged here.

10/29/2023