Machine Learning Adventures

Hello! This is a collection of experiments, re-implmentations, and fun things I have done in machine learning. All of the code will be written in Python and I commonly use the libraries numpy, keras, and pytorch

Folders

Each of the folders contains the code for creating a deep learning model for that dataset. All the files should have comments explaning the overall thought process. Some folders may have pre-trained weights included. I plan on attatching accuracies and graphs later on.

Current Folders

• kddcup1999 - A cybersecurity dataset which can be found here
  • Feed-forward neural network with categorical and continuous features
  • Embedding layers utilized in PyTorch implementation and one-hot encoding used in Keras implementation
• wisdm - A human activity dataset which can be found here
  • Recurrent neural network on time series [lstm]
  • Implements sliding window to group data into trainable information
  • Inspired by Tomasz Bartkowiak's implementation
• annotated_transformer - A re-implmentation of the transformer for language translation
  • Most of the code is from the annotated transformer paper
  • Implements both the encoder and decoder stacks with additional comments
  • Integrates some of the illustrations from the illustrated transformer

Useful resources

• Overall
  • PyTorch Basics - MorvanZhou
  • PyTorch Basics - yunjey
  • Overview of loss functions
  • Save memory with pandas
• RNN (LSTM)
  • Basics
  • LSTM Input
  • Sliding window for time series
• Transformers
  • Annotated Transformer
  • Illustrated Transformer