Please enable JavaScript.

Coggle requires JavaScript to display documents.

Model Data (Feature understanding and selection (Data types (binary, multi…

- - - - Hover, missing values
      - Unique column
      - mean, median std dev, etc
        
        select number of bins
        
        (1,3) =2
        
        (1,3] = 2,3
        
        [1,3) = 1,2
        
        [1,3] = 1,2,3
  - - - Frequent values table
      - AutoML would not know which to remove
  - - - Regressions, neural networks, support vector machines
      - Nulls
      - Joins
        
        Not Applicable
- - - - Log Loss - accuracy
  - - - cross validation and hold out samples
    - - "n" folds
        
        small sets ---> more folds
        
        Cross validation
    - - train, validation, holdout
    - - Allows for specification of group membership
      - DataRobot decides partition
  - - - abandoning shopping carts
      - Quick run feature
        
        Abbreivated version of autopilot
        
        16% vs 32%
    - - Data robot is updating too fast for this book
    - - Begin with 500 mb of data
      - Importance of scores calculated
      - Calculate list models
    - - One feature against all other features
      - Non-linear relationships
  - - - dedicated computer
    - - "models"
        
        LogLoss measures bad loss
        
        Decision tree will not be used
        
        Gradient based models
      - Random forest qualifier
        
        python based models
        
        Gini impurity
        
        Which models survive?
- - - - Raw, uncleaned data
      - Start small
      - csv okay
    - - Feature names unique
        
        remove or rename it
      - upload to datarobot cloud
      - "patient decisions"
        
        "data" blue, data currently viewed