In order to better understand different machine learning & deep learning models, I try to implement models from scratch. Complex models might directly use pytorch implementation of simple functions (For example, this repo will not repeated implement or reference linear layer in PyTorch).
Try to implement all traditional machine learning models with consistent format, now the models follow .fit() and .predict() functions to train and predict like SKLearn.
For deep learning models, I use PyTorch to avoid the calculation of back-prop and derivatives. All implemented models are directly to use like any other PyTorch models.
PyTorchNumPy