Perfectly Awesome

Part 1 (Intro):

• intro
• source code
• PyTest
• dvc
• Anaconda environments (for isolating project dependencies)

Part 2 (Data acquisition / analysis):

• blog post
• Liar Liar dataset
• Liar Liar, augmented

• Pandas

• Analyze label distributions
• Ngram analysis
  • TF-IDF weightings
• Topic modeling
  • LDA (Wikipedia)
  • LSA (Wikipedia)
• Sentiment analysis
  • (compare sentiment score between True & False statements)
  • VADER algorithm

Part 3 (First model):

• Model V1
  • Preprocessing
    • cleaner script
  • Model definition
    • random forest as baseline
    • definition (github)
  • Features
    • Use 2 types: manual and ngram
    • one-hot encoding
  • Training Pipeline
    • json format
    • yaml format
    • jsonnet format
    • pipeline code boilerplate
    • MLFlow tracking API
  • Functionality tests
    • testing (Jeremy Jordan)
    • testing (Eugene Yan)
    • testing (@keeper6928)
    • normalization tests
    • modeling tests
    • “Great Expectations” library
    • testing script
  • Wrapup

Part 4 (refined model):

• Model V2

  • Feature Importance
    • Gini impurity
    • error analysis
    • Shapely values
    • SHAP library
  • Error Analysis
  • What Does the Data Say?
  • Model V2
    • Transformer model
    • Roberta

Part 5 (Deployment):

• Prediction Rest API
• Chrome Extension
• Continuous Integration (CI)