// Slices the data into sections, returning a slice of float64s.
// Returns an error if the data set is empty.
func (q Quartile) Transform(analyzer *mino.Analyzer, data mino.Collection) (interface{}, error) {
sections := q.Sections
if sections == 0 {
sections = 4
}
results := make([]float64, sections-1)
dividend := (1 + data.Weight()) / float64(sections)
// todo(connor4312): my algorithm below behaves oddly when the data
// size is less than the number of requested sections. Whoever discovers
// a solution to it is awarded three cookies.
if data.Len() < sections {
return results, InsufficientDataError
}
data.Sort(func(a, b mino.DataPoint) bool {
return a.Value < b.Value
})
var weight float64
var prev mino.DataPoint
SECTION:
for i := 1; i < sections; i++ {
target := float64(i) * dividend
for data.HasMore() {
current := data.Next()
last := prev
prev = current
if weight+current.Weight > target {
frac := (target - weight) / current.Weight
results[i-1] = last.Value*(1-frac) + current.Value*frac
data.Rewind(1)
continue SECTION
}
weight += current.Weight
}
results[i-1] = prev.Value
}
return results, nil
}
// Performs a standard deviation, and returns StddevResults. If the data
// set's length is less than 2, then an error will be returned along
// with results of a standard deviation and average of zero.
// These results may be inspected, however some
// methods (such as LogProb) will fail.
//
// Based on the algorithm from RosettaCode, which can be found at:
// http://rosettacode.org/wiki/Standard_deviation#Go, and
// adapted for weighted standard deviation, the equation
// for which which equation which was found at:
// http://www.itl.nist.gov/div898/software/dataplot/refman2/ch2/weightsd.pdf
func (s Stddev) Transform(analyzer *mino.Analyzer, data mino.Collection) (interface{}, error) {
l := data.Len()
// Special case: not enough data in set
if l == 0 {
return StddevResults{0, 0}, InsufficientDataError
} else if l == 1 {
return StddevResults{data.Next().Value, 0}, InsufficientDataError
}
var average, variance float64
n := float64(0)
for data.HasMore() {
x := data.Next()
n += x.Weight
a := average + (x.Value-average)/(n/x.Weight)
variance, average = variance+(x.Value-average)*(x.Value-a)*x.Weight, a
}
besselLen := data.Len()
if s.Bessel {
besselLen -= 1
}
return StddevResults{
Average: average,
Deviation: math.Sqrt((float64(data.Len()) * variance) /
(float64(besselLen) * data.Weight())),
}, nil
}