The repository is a collection of my self-taught notebooks for data science theories and practices. A huge effort is made to strike a balance between methodology derivation (with math) and hands-on coding. The target audience is data science practitioners (including myself) with hands-on experiences who are seeking for more in-depth understandings of machine learning algorithms and relevant statistics.

Here to visit the web site Hello, Data Science! hosting all the notebooks in nicely rendered HTML.

notebooks/

A notebook is written in either Jupyter or R markdown. The major programming languages used for most of the notebooks are Python and/or R. You may find me sometimes inter-operate the two langauges in a single notebook. This is achieved thanks to reticulate.

• Statistics
  • Bayesian Modeling Explained
  • Bootstrap Sampling 101
  • Introduction to Topic Models
  • A Gentle Walkthrough of the Economist's Excess Deaths Model for COVID-19
• Machine Learning
  • Neural Networks Fundamentals
  • Matrix Factorization for Recommender Systems
  • Introduction to Learning-to-Rank
  • General-Purpose Ranking Models
  • On Model Explainability
  • Demystify Modern Gradient Boosting Trees
• Natural Language Understanding
  • On Subword Units
  • [Contex-Free Word Embeddings]
  • [Contex-Aware Word Embeddings]
• Data Engineering
  • Infrastructure-as-Code: A Terraform AWS Use Case
  • Serverless Deployment: AWS Lambda with HTTP API
• Programming
  • R
    • Asynchronous Programming in R
    • High Performance Computing in R using Rcpp
• Projects
  • YouTube-8M Multi-Label Video Classification
  • Visualization for Taiwanese General Election 2020
  • A Note on Model Evaluation for Imbalanced Data
  • A Short Ride on AWS DeepRacer 2020
  • A Simulation Exercise for Your PS5 Lottery Draws

labs/

These are quick-and-dirty scripts to explore a variety of open source machine learning tools. They may not be completed and can be messy to read.

To ensure reproducibility it is recommended to use pyenv along with pyenv-virtualenv to control both Python and package version.

pyenv support only Linux and macOS. For Windows user it is recommended to use conda instead.

To use virtualenv with reticulate in Rmd, the involved Python must be installed with shared library:

PYTHON_CONFIGURE_OPTS="--enable-shared" pyenv install 3.7.0

Each notebook has different package dependencies. Here is an example to create an environment specific for the notebook on model explainability:

cd notebooks/ml/model_explain
pyenv virtualenv 3.7.0 k9-model-explain
pyenv local k9-model-explain
pip install --upgrade pip
pip install -r requirements.txt

• Machine Learning
  • Factorization Machines
  • Recurrent Neural Nets
  • Sequence-to-Sequence Models
  • GANs
  • Reinforcement Learning Basics
  • Approximated Nearest Neighbor
• Statistics
  • Law of Large Numbers and Central Limit Theorem
  • On Linear Regression: Machine Learning vs Econometrics
  • Linear Mixed Effects Models
  • Naive Bayes
  • Bayesian Model Diagnostic
  • Bayesian Time Series Forecasting
• Tools/Programming
  • PyTorch Hands-On
  • RASA Chatbot Framework Hands-On
• Programming
  • R
    • Production Quality Shiny App Development
  • Python
    • Dash for Interactive Dashboarding
• Projects
  • Model Deployment with gRRC

• Dockerize each notebook (for complete reproducibility and portability)?
• Tidy up dependencies for each notebook