Please enable JavaScript.

Coggle requires JavaScript to display documents.

Chapters 13-15 (Chapter 14 - Feature Understanding Selection (Data types…

- - - - Regression
      - Neural networks
      - Support vector machines
- - - - Model is evaluated on probabilities generated and difference from the correct answer
      - Ex: Decides to use TF (0 or 1) as T = .1 and F = .9 as a potential answer
    - - Random
      - Stratified
        
        Tries to maintain distribution of samples in holdout sample
      - 5 and 10 are common values for number of folds
    - - You decide exactly which data to use for each fold
      - Train, validation, and holdout samples manually specified
        
        Helpful to make sure there is no overlap for example between friendship groups
    - - Like partition, but keeps group data together throughout process
  - - - Starts with 16% data sample and chooses best model by 32% data sample
    - - Starts with the 32% data sample and models that have performed well historically
    - - Relative importance of a particular feature in the prediction target
  - - - If one model is good at predicting certain features, and another good at other features, then which model is best will be used determined on the prediction target and available data