This is a simple neural network that teaches itself to recognize handwritten numbers. To train and test the network, data from the MNIST database was used.
https://de.wikipedia.org/wiki/MNIST-Datenbank
Most of the knowledge about the program comes from the wonderful book: „Make Your Own Neural Network“ from Tariq Rashid
And by the way, i prefer PyCharm from JetBrains over VS Code for Python-Projects, when i deal with machine learing next to Anaconda and Jupyter Notebooks.
(althougt i also often use VS Code) because:
- many things just work automatically without big settings and plugins (git, conection to github, venv, import Packages in the project, SciView for plots and data)
- my students and my dauther also get along well with it
2018 must have been when I wanted to understand roughly how neural networks work. During my research I came across the fantastic book mentioned above.
I worked through the book intensively, but didn't write a line of Python code on the topic myself. Inspired by the book, I went deeper into the mathematics behind machine learning and read various math books and solved some simple problems.
Over the past few months, I've been reading the following books on and off:
- „Neuronale Netze Programmieren mit Python“ von Joachim Steinwender und Roland Schwaiger
- „Neural Networks from Scratch in Python“ from Harrison Ainsley and Daniel Kukiela
But this book is not easy for me.
- „Mathematics for Machine Learning“ from Marc Peter Deisenroth, A. Aldo Faisal and Cheng Soon One
And so it happened that I programmed my first neural network with Python myself in the last week.
If only one had more time to learn... its a technically fascinating time.
- try useful values/settings for hiddenNotes and the learning rate automatically
- save which values give the best results
- save the network
- value photos from your own handwriting
- train and test the programm with the zalando database https://github.com/zalandoresearch/fashion-mnist