var trainingSet dataset.Dataset
BeforeEach(func() {
var err error
estimator, err = gradientdescentestimator.NewGradientDescentParameterEstimator(
0.000001,
0.001,
50,
lossGradient,
)
Ω(err).ShouldNot(HaveOccurred())
})
Context("When the estimator hasn't been trained", func() {
It("Returns an error", func() {
_, err := estimator.Estimate([]float64{0.02, 0.1, 0.1, 0.1})
Ω(err).Should(BeAssignableToTypeOf(gdeErrors.UntrainedEstimatorError{}))
})
})
Context("When the parameter guess is empty", func() {
BeforeEach(func() {
err := estimator.Train(trainingSet)
Ω(err).ShouldNot(HaveOccurred())
})
It("Returns an error", func() {
_, err := estimator.Estimate([]float64{})
Ω(err).Should(BeAssignableToTypeOf(gdeErrors.EmptyInitialParametersError{}))
})
})
err = trainingSet.AddRowFromStrings([]string{
fmt.Sprintf("%.10f", x0),
fmt.Sprintf("%.10f", x1),
fmt.Sprintf("%.10f", y),
})
Ω(err).ShouldNot(HaveOccurred())
}
}
err = estimator.Train(trainingSet)
Ω(err).ShouldNot(HaveOccurred())
})
Context("Given a mis-shaped initial parameter guess", func() {
It("Returns an error", func() {
_, err := estimator.Estimate([]float64{1.75, -3.25})
Ω(err).Should(HaveOccurred())
})
})
Context("Given a valid initial parameter guess", func() {
It("Does not return an error", func() {
_, err := estimator.Estimate([]float64{1.75, -3.25, 4.25})
Ω(err).ShouldNot(HaveOccurred())
})
It("Estimates the true parameters", func() {
estimatedParameters, _ := estimator.Estimate([]float64{1.75, -3.25, 4.25})
Ω(estimatedParameters).Should(HaveLen(3))
for i := 0; i < 3; i++ {
