Regression

• Regression:
  • Better for inputs that are changing continually through time
  • Computes numerical value from inputs
  • Outputs change smoothly as the inputs change
  • Used to control continuous parameters like volume, position, etc… 
  • Most common machine learning technique in the development of new electronic instruments
  • Unlike classification, regression models can output values that are not in the training set
  • Values in between or outside of the values in the training set

• Good Uses in Music:
  • Raise pitch as I move from position X to position Y
  • Decrease amount of drums as lights become dimmer
  • Match tempo of the music to tempo of my body movements

Examples of Types of Regression-

• Least Squares Linear Regression
  • Linear Regression Model will find an equation to fit the set
    • 1 input: y = ax+b
    • 2 inputs: y=ax+bx+c
    • etc… 
  • Programming coordinate graphs use y=0 at top of the screen and y increasing as you go down
    • Slope negative/positive switch in equation (not in visuals)
  • Linear Regressions Algorithm WILL produce a straight (linear) line to fit the set; do not use with quadratic/exponentials sets
  • Good for controlling volume or pitch in instruments where you only want increase/decrease linearly (no curve)

• Polynomial Regression
  • Need to set up to deal with the appropriate polynomial degree
  • Can kinda fit a polynomial to a straight line but will be slightly curvy still

• When to use which model?
  • Linear Regression Lines will do better at ignoring outliers
  • Can always edit training data to improve your model