Please enable JavaScript.

Coggle requires JavaScript to display documents.

Larson Chapter 15 (Model Selection Process (Each different algorithm is…

- - - - Some will need up to 10 CPUs
        
        In example, 16% of data is being used to train models which will eventually run 64% of data
        
        Think of this as DataRobot running a Tournament to determine the best algorithm
  - - - Check models tab to view completed models so far
        
        The distinction between algorithm performance in the leaderboard is important because it is where DataRobot will select the best models from
  - - - Line also indicates the Blueprint used (i.e. BP3) for this model
        
        Model number is denoted too (i.e. M6)
        
        Beta symbol will indicate positive or negative coefficients
- - - - LogLoss trains the model based on its accuracy (ex. how close its prediction is to actual outcome of True/False
- - - - Specifically, QuickRun jumps straight to using 32% of the sample to run models that have historically performed well
      - Running with AutoPilot
        
        Rapid secession of 7 steps
        
        1.) Set Target Feature
        
        2.) Creating CV and Holdout Partition
        
        3.) Characterizing Target Variable
        
        4.) Loading Dataset
        
        1 more item...
        
        Output will yield several columns
        
        Green bar indicates relative importance of model
        
        Uses ACE number which is equivalent to R^2 scores
- - - - Blender tool will do this
      - Blending models will select and output predictions based on the best models