Lab 4, Part 1 - data-ppf/data-ppf.github.io GitHub Wiki

Bouk and 'smoothing' mortality curves

context

• hypothesis of age being the most important activity
• random geocities website for data
• activity: 4 minutes to find birth / death statistics from random website

example: obtaining, scrubbing, exploring, modeling, interpreting

• obtaining & scrubbing data: example of real-world data analysis

  • using lynx, awk, grep on data set
  • problems of pdfs
  • problems of time zones to be watched for
  • lots of shellscript fu on web sites
  • hypothesis of singers vs. poets

• exploratory data analysis

  • load
  • compute probability
  • start plotting
  • some basic intuitions around insurance
  • hazard, survival probability
  • then can start making money
  • calculate survival (product of probs)

• modeling the data

  • straight line as model/design choice
  • many procedures to draw trend lines
  • subjective design choices
  • example of dangers of complicated models
  • "negative consequences of choosing a crazy model"

• interpreting / goals

  • example of herschel not clear picking curve of sinuosity
  • marketing choice: beauty
  • making money
  • boss like curve wiggly
  • "modern machine learning": prediction primary goal
  • alternate: maximize success metric (e.g., profit, sales...)