// CompareOutOfSample will compare the current hypothesis function learn by linear regression whith respect to 'f' out of sample
func (linreg *LinearRegression) CompareOutOfSample(f linear.LinearFunc, nParams int) float64 {
outOfSample := 1000
diff := 0
for i := 0; i < outOfSample; i++ {
//var oY int
oX := make([]float64, linreg.VectorSize)
oX[0] = float64(1)
for j := 1; j < len(oX); j++ {
oX[j] = linreg.Interval.RandFloat()
}
gi := float64(0)
for j := 0; j < len(oX); j++ {
gi += oX[j] * linreg.Wn[j]
}
if nParams == 2 {
if linear.Sign(gi) != int(f(oX[1], oX[2])) {
diff++
}
} else if nParams == 1 {
if linear.Sign(gi) != int(f(oX[1])) {
diff++
}
}
}
return float64(diff) / float64(outOfSample)
}
// CompareInSample will compare the current hypothesis function learn by linear regression whith respect to 'f'
func (linreg *LinearRegression) CompareInSample(f linear.LinearFunc, nParams int) float64 {
gInSample := make([]float64, len(linreg.Xn))
fInSample := make([]float64, len(linreg.Xn))
for i := 0; i < len(linreg.Xn); i++ {
gi := float64(0)
for j := 0; j < len(linreg.Xn[0]); j++ {
gi += linreg.Xn[i][j] * linreg.Wn[j]
}
gInSample[i] = float64(linear.Sign(gi))
if nParams == 2 {
fInSample[i] = f(linreg.Xn[i][1], linreg.Xn[i][2])
} else if nParams == 1 {
fInSample[i] = f(linreg.Xn[i][1])
}
}
// measure difference:
diff := 0
for i := 0; i < len(linreg.Xn); i++ {
if gInSample[i] != fInSample[i] {
diff++
}
}
return float64(diff) / float64(len(linreg.Xn))
}
// EoutFromFile only supports linear regressions with transformed data.
// todo:(santiaago) make this more generic.
func (linreg *LinearRegression) EoutFromFile(filename string) (float64, error) {
file, err := os.Open(filename)
if err != nil {
log.Fatal(err)
}
defer file.Close()
numError := 0
numberOfLines := 0
scanner := bufio.NewScanner(file)
for scanner.Scan() {
split := strings.Split(scanner.Text(), " ")
var line []string
for _, s := range split {
cell := strings.Replace(s, " ", "", -1)
if len(cell) > 0 {
line = append(line, cell)
}
}
var oY int
var oX1, oX2 float64
if x1, err := strconv.ParseFloat(line[0], 64); err != nil {
fmt.Printf("x1 unable to parse line %d in file %s\n", numberOfLines, filename)
return 0, err
} else {
oX1 = x1
}
if x2, err := strconv.ParseFloat(line[1], 64); err != nil {
fmt.Printf("x2 unable to parse line %d in file %s\n", numberOfLines, filename)
return 0, err
} else {
oX2 = x2
}
oX := linreg.TransformFunction([]float64{float64(1), oX1, oX2})
if y, err := strconv.ParseFloat(line[2], 64); err != nil {
fmt.Printf("y unable to parse line %d in file %s\n", numberOfLines, filename)
return 0, err
} else {
oY = int(y)
}
gi := float64(0)
for j := 0; j < len(oX); j++ {
gi += oX[j] * linreg.Wn[j]
}
if linear.Sign(gi) != oY {
numError++
}
numberOfLines++
}
if err := scanner.Err(); err != nil {
log.Fatal(err)
}
return float64(numError) / float64(numberOfLines), nil
}
// Hypothesis function h is the hypothesis of the perceptron algorithm.
// It takes as arguments vector X and a vector W (w for weight)
// function h implements h(x) = sign(w'x)
func h(x Point, w Point) int {
if len(x) != len(w) {
fmt.Println("Panic: vectors x and w should be of same size.")
panic(x)
}
var res float64 = 0
for i := 0; i < len(w); i++ {
res = res + w[i]*x[i]
}
return linear.Sign(res)
}
func (linreg *LinearRegression) EValIn() float64 {
gInSample := make([]int, len(linreg.XVal))
for i := 0; i < len(linreg.XVal); i++ {
gi := float64(0)
for j := 0; j < len(linreg.XVal[0]); j++ {
gi += linreg.XVal[i][j] * linreg.Wn[j]
}
gInSample[i] = linear.Sign(gi)
}
nEin := 0
for i := 0; i < len(gInSample); i++ {
if gInSample[i] != linreg.YVal[i] {
nEin++
}
}
return float64(nEin) / float64(len(gInSample))
}
// Eout is the fraction of out of sample points which got misclassified.
func (linreg *LinearRegression) Eout() float64 {
outOfSample := 1000
numError := 0
for i := 0; i < outOfSample; i++ {
var oY int
oX := make([]float64, linreg.VectorSize)
oX[0] = float64(1)
for j := 1; j < len(oX); j++ {
oX[j] = linreg.Interval.RandFloat()
}
flip := 1
if linreg.Noise != 0 {
r := rand.New(rand.NewSource(time.Now().UnixNano()))
rN := r.Intn(100)
if rN < int(math.Ceil(linreg.Noise*100)) {
flip = -1
}
}
// output with potential noise in 'flip' variable
if !linreg.TwoParams {
oY = evaluate(linreg.TargetFunction, oX) * flip
} else {
oY = evaluateTwoParams(linreg.TargetFunction, oX) * flip
}
gi := float64(0)
for j := 0; j < len(oX); j++ {
gi += oX[j] * linreg.Wn[j]
}
if linear.Sign(gi) != oY {
numError++
}
}
return float64(numError) / float64(outOfSample)
}