regression.nim

# not sure why this isn't built-in
template inc(self: var float; other: float) =
  self = self + other

type
  Averager* = object
    count: int
    x: float
 
proc feed*(self: var Averager; value: float) =
  inc self.x, value
  inc self.count

proc done*(self: var Averager): float =
  result = self.x / self.count.float
  self.count = 0
  self.x = 0

type
  Deviator* = object
    count: int
    x: float
 
proc feed*(self: var Deviator; value, mean: float) =
  let val = (value - mean)
  inc self.x, pow(val, 2)
  inc self.count

proc done*(self: var Deviator; population: bool = true): float =
  if population:
    result = self.x / self.count.float
  else:
    result = self.x / (self.count - 1).float
  self.count = 0
  self.x = 0

type
  Correlator* = object
    ## Computes Pearson R from a stream of x/y values.
    xy, x2, y2: float
 
proc feed*(self: var Correlator; x, y: float) =
  inc self.xy, x * y
  inc self.x2, pow(x, 2)
  inc self.y2, pow(y, 2)

proc done*(self: var Correlator): float =
  result = self.xy / sqrt(self.x2 * self.y2)
  self.xy = 0
  self.x2 = 0
  self.y2 = 0

proc linear_regression*(r, mx, my, sx, sy: float; slope, intercept: out float) =
  ## Computes the slope and intercept for a linear regression line given correlation R, means of x and y, and standard deviations of x and y.
  slope = r * sy / sx
  intercept = my - slope * mx