Please enable JavaScript.

Coggle requires JavaScript to display documents.

Chapters 13-15 (DataRobot (Always inspect every single variable - content,…

- - - - Each column is labelled and selectable
        
        Dropdown menus include graphs, bar charts.
  - - - Correct Categorical features if they are set to Numeric
- - - - 80-20 is a typical split. Also there are 4 different partition methods
- - - - Does the unit reappear?
      - Find first instance of unit
      - Order the cases by time
      - End time
      - Beginning time
  - - - Prediction speed
      - Speed to build
      - Familiarity with model
      - Insights
      - Predictive accuracy
- - - - This is still counted as textual or categorical.
        
        Some algorithms will SKIP rows that have ANY missing values
        
        Such as: Regression, neural networks, and support
        vector machines.
        
        Check for human-entered missing vals like:
        
        NULL, null, N/A, #N/A, n/a, etc
        
        Or spaces, blanks, "None"
        
        Or INF, inf, infinite
- - - - Filetypes
        
        ODBC
        
        HDFS
        
        Can upload from URL
        
        CSV files, comma split
        
        Can accept compressed ZIP files
        
        Choose a file
      - EX
        
        A managed healthcare organization believes they suffer from over $30
        million in preventable losses annually due to the readmission of
        patients who are discharged from the hospital too soon. However,
        keeping all patients in the hospital longer is costly and inconvenient to
        patients. The organization has hired you to develop a model of
        readmission risk that doctors can consider when determining when to
        discharge a patient (if you find anything else in the data that helps
        reduce readmissions, they will appreciate that too).
        They want you to start by focusing on readmissions among their
        diabetic patient population, and they have provided a dataset
        (10kDiabetes.csv) from a large set of hospitals, including their own,
        which describes admissions of this patient population. Predicted
        readmission risks from the model are only useful if the doctors believe
        in you and your model. To help the doctors gain trust, you have been
        invited to create a presentation for the doctors. Sketch out the content
        of this report or whitepaper. Focus on the conclusions and exhibits you
        will show them to support those conclusions.