func GradientDescent(
initialGuess []float64,
learningRate, precision float64,
maxIterations int,
gradient func([]float64) ([]float64, error),
) ([]float64, error) {
if len(initialGuess) == 0 {
return nil, errors.New("initialGuess cannot be empty")
}
oldResult := make([]float64, len(initialGuess))
newResult := make([]float64, len(initialGuess))
copy(oldResult, initialGuess)
for i := 0; i < maxIterations; i++ {
gradientAtOldResult, err := gradient(oldResult)
if err != nil {
return nil, err
}
newResult = vectorutilities.Add(oldResult, vectorutilities.Scale(-learningRate, gradientAtOldResult))
if (knnutilities.Euclidean(newResult, oldResult, precision)) < precision*precision {
return newResult, nil
} else {
oldResult = newResult
}
}
return newResult, nil
}
package vectorutilities_test
import (
"github.com/amitkgupta/goodlearn/vectorutilities"
. "github.com/onsi/ginkgo"
. "github.com/onsi/gomega"
)
var _ = Describe("Vector Utilities", func() {
Describe("Add", func() {
It("Adds", func() {
x := []float64{1.01, 2, 3, 4}
y := []float64{3.14, -1, 0, 0}
Ω(vectorutilities.Add(x, y)).Should(Equal([]float64{4.15, 1, 3, 4}))
})
})
Describe("Scale", func() {
It("Scales", func() {
x := []float64{1, 2, 3, 4}
a := -2.33
Ω(vectorutilities.Scale(a, x)).Should(Equal([]float64{-2.33, -4.66, -6.99, -9.32}))
})
})
})