Please enable JavaScript.

Coggle requires JavaScript to display documents.

CHAPTERS 17&18 (EVALUATE MODEL PERFORMANCE (decision tree algorithm:…

- - - - explains percentage of target
  - - - left: what car SHOULD HAVE seen
      - top: actual detection made by the in-car algorithm
      - TRUE POSITIVE (TP): the light is
        green, and the algorithm has determined that it is green
      - TRUE NEGATIVE (TN):the light is red, and the algorithm has determined that it is red
      - FALSE POSITIVE (FP): traffic light was red, but the algorithm classified it as green
        
        causes car to get t-boned
      - FALSE NEGATIVE (FN): the traffic light was green, while the algorithm determined that the light is red.
        
        stops quickly, getting rear ended
      - every case is evaluated (validated) and placed in one of these four quadrants
      - "distinction between true positive and true negative is simply whatever the system decided constituted a positive"
- - - - true positive rate and the false positive rate are zero
        
        “curve” remains there until the model begins predicting negative cases
    - - screen displaying which model is right when two models disagree
        
        calculated by subtracting the predictions made by the two models
        
        ordered from lowest to highest and grouped into 15 bins with the smallest (negative) numbers to the left
      - orange line shows which model is closest to the average of the readmissions for the bin
  - - - how long it takes to train a model
    - - able to understand the exact meaning of its results
    - - different algorithms make different statistics available