Machine learning day - lecture notes

Notes from Praseed Pai’s lecture, Machine learning day - KMUG (Praseed Pai) - 9-JUL-2016

• Analytical thinking vs system thinking
  • analytical
    • break problem down and solve
  • system thinking
    • holistic approach, nonlinear
    • assume dependent vars

• Algorithmic techniques

  • Hilbert space methods
    • proximity queries between datasets
    • Hilbert’s 23 problems (esp. pbm #2 and #10)
  • statistical learning
    • types of statistics
      • non parametric statistics
        • categorical/nominal data
        • ordinal data (signifies order)
      • parametric
        • ratio
        • interval
    • descriptive
      • central tendency
      • dispersion
      • association
  • deep learning
    • neural networks

• Algorithmic classification

  • supervised learning
    • classification
    • regression/prediction
      • classification based on numerical data
  • unsupervised
    • clustering
    • dimensionality reduction
  • association analysis
    • apriori

      • eg. Given historical retail data, decide whether customers who purchase bread and sugar should be offered a coupon for another product, say beer. We can solve this by finding the % of baskets that have beer in addn. to bread and sugar.

      • P(Y	X) - prob of Y given X.

          TxnID  | items
        ---------+--------------------
          1      | shoes, shirt, jacket
          2      | shoes, jacket
          3      | shoes, jeans
          4      | shirt, sweatshirt
        
          items           | Frequency
        ------------------+----------
          shoes           | 75%
          shirt           | 50%
          {shoes, jacket} | 50%
        

• Decision tree classifier
  • generate decision tree based on inputs
• Naive Bayes
  • initial condition - priori probability
  • adjust probability based on new data
  • assumes independent variables
  • posterior probability - calculate based on priori data
  • eg. given learning data height, weight, foot size, predict gender.
  • read up:
    • false positive % - sensitivity
    • false negative % - specificity
    • base rate
    • monty hall problem
    • Weka - ML tool

MOOCs

• Caltech ML course by yasser abu mustafa
• Weka MOOC

Recommended books

• Machine Learning (Tom Mitchell)
• Statistics hacks (Bruce Frey)
• Financial Numerical Recipes in C++ (available online)
• Data Science for Dummies Using Python
• Machine Learning with Scikit Learn - Packt