Projects with neural networks programmed by scratch in Numpy
3 different proyects developed for a Universidad de Buenos Aires course in 2023.
Two different Neural Networks where programmed only using numpy. The NN are objects which can be instanciated, trained and retrieve their parameters and update lists. Also trained objects can be used to predict.
The challenge was to develop everything from scratch and only using Numpy in an (more or less) efficient way.
The supervised NN consist in a classic multilayered approach, while the unsupervised explores the modified Oja and Sanger approach.