Golang math.Inf函数代码示例

// NewHeatMap creates as new heat map plotter for the given data,
// using the provided palette. If g has Min and Max methods that return
// a float, those returned values are used to set the respective HeatMap
// fields.
func NewHeatMap(g GridXYZ, p palette.Palette) *HeatMap {
	var min, max float64
	type minMaxer interface {
		Min() float64
		Max() float64
	}
	switch g := g.(type) {
	case minMaxer:
		min, max = g.Min(), g.Max()
	default:
		min, max = math.Inf(1), math.Inf(-1)
		c, r := g.Dims()
		for i := 0; i < c; i++ {
			for j := 0; j < r; j++ {
				v := g.Z(i, j)
				if math.IsNaN(v) {
					continue
				}
				min = math.Min(min, v)
				max = math.Max(max, v)
			}
		}
	}

	return &HeatMap{
		GridXYZ: g,
		Palette: p,
		Min:     min,
		Max:     max,
	}
}

// TestFloatCmpSpecialValues tests that Cmp produces the correct results for
// combinations of zero (±0), finite (±1 and ±2.71828), and infinite (±Inf)
// operands.
func TestFloatCmpSpecialValues(t *testing.T) {
	zero := 0.0
	args := []float64{math.Inf(-1), -2.71828, -1, -zero, zero, 1, 2.71828, math.Inf(1)}
	xx := new(Float)
	yy := new(Float)
	for i := 0; i < 4; i++ {
		for _, x := range args {
			xx.SetFloat64(x)
			// check conversion is correct
			// (no need to do this for y, since we see exactly the
			// same values there)
			if got, acc := xx.Float64(); got != x || acc != Exact {
				t.Errorf("Float(%g) == %g (%s)", x, got, acc)
			}
			for _, y := range args {
				yy.SetFloat64(y)
				got := xx.Cmp(yy)
				want := 0
				switch {
				case x < y:
					want = -1
				case x > y:
					want = +1
				}
				if got != want {
					t.Errorf("(%g).Cmp(%g) = %s; want %s", x, y, got, want)
				}
			}
		}
	}
}

// Train computes and stores the bin values
// for the training instances.
func (b *BinningFilter) Train() error {

	as := b.getAttributeSpecs()
	// Set up the AttributeSpecs, and values
	for attr := range b.attrs {
		if !b.attrs[attr] {
			continue
		}
		b.minVals[attr] = float64(math.Inf(1))
		b.maxVals[attr] = float64(math.Inf(-1))
	}

	err := b.train.MapOverRows(as, func(row [][]byte, rowNo int) (bool, error) {
		for i, a := range row {
			attr := as[i].GetAttribute()
			attrf := attr.(*base.FloatAttribute)
			val := float64(attrf.GetFloatFromSysVal(a))
			if val > b.maxVals[attr] {
				b.maxVals[attr] = val
			}
			if val < b.minVals[attr] {
				b.minVals[attr] = val
			}
		}
		return true, nil
	})

	if err != nil {
		return fmt.Errorf("Training error: %s", err)
	}
	b.trained = true
	return nil
}

// generateValidatedLengthExample generates a random size array of examples based on what's given.
func (eg *exampleGenerator) generateValidatedLengthExample() interface{} {
	minlength, maxlength := math.Inf(1), math.Inf(-1)
	if eg.a.Validation != nil {
		if eg.a.Validation.MinLength != nil {
			minlength = float64(*eg.a.Validation.MinLength)
		}
		if eg.a.Validation.MaxLength != nil {
			minlength = float64(*eg.a.Validation.MaxLength)
		}
	}
	count := 0
	if math.IsInf(minlength, 1) {
		count = int(maxlength) - (eg.r.Int() % 3)
	} else if math.IsInf(maxlength, -1) {
		count = int(minlength) + (eg.r.Int() % 3)
	} else if minlength < maxlength {
		count = int(minlength) + (eg.r.Int() % int(maxlength-minlength))
	} else if minlength == maxlength {
		count = int(minlength)
	} else {
		panic("Validation: MinLength > MaxLength")
	}
	if !eg.a.Type.IsArray() {
		return eg.r.faker.Characters(count)
	}
	res := make([]interface{}, count)
	for i := 0; i < count; i++ {
		res[i] = eg.a.Type.ToArray().ElemType.GenerateExample(eg.r)
	}
	return res
}

// checkIsBestApprox checks that f is the best possible float64
// approximation of r.
// Returns true on success.
func checkIsBestApprox(t *testing.T, f float64, r *Rat) bool {
	if math.Abs(f) >= math.MaxFloat64 {
		// Cannot check +Inf, -Inf, nor the float next to them (MaxFloat64).
		// But we have tests for these special cases.
		return true
	}

	// r must be strictly between f0 and f1, the floats bracketing f.
	f0 := math.Nextafter(f, math.Inf(-1))
	f1 := math.Nextafter(f, math.Inf(+1))

	// For f to be correct, r must be closer to f than to f0 or f1.
	df := delta(r, f)
	df0 := delta(r, f0)
	df1 := delta(r, f1)
	if df.Cmp(df0) > 0 {
		t.Errorf("Rat(%v).Float64() = %g (%b), but previous float64 %g (%b) is closer", r, f, f, f0, f0)
		return false
	}
	if df.Cmp(df1) > 0 {
		t.Errorf("Rat(%v).Float64() = %g (%b), but next float64 %g (%b) is closer", r, f, f, f1, f1)
		return false
	}
	if df.Cmp(df0) == 0 && !isEven(f) {
		t.Errorf("Rat(%v).Float64() = %g (%b); halfway should have rounded to %g (%b) instead", r, f, f, f0, f0)
		return false
	}
	if df.Cmp(df1) == 0 && !isEven(f) {
		t.Errorf("Rat(%v).Float64() = %g (%b); halfway should have rounded to %g (%b) instead", r, f, f, f1, f1)
		return false
	}
	return true
}

func TestAllSetDefaults(t *testing.T) {
	// Exercise SetDefaults with all scalar field types.
	m := &Defaults{
		// NaN != NaN, so override that here.
		F_Nan: Float32(1.7),
	}
	expected := &Defaults{
		F_Bool:    Bool(true),
		F_Int32:   Int32(32),
		F_Int64:   Int64(64),
		F_Fixed32: Uint32(320),
		F_Fixed64: Uint64(640),
		F_Uint32:  Uint32(3200),
		F_Uint64:  Uint64(6400),
		F_Float:   Float32(314159),
		F_Double:  Float64(271828),
		F_String:  String(`hello, "world!"` + "\n"),
		F_Bytes:   []byte("Bignose"),
		F_Sint32:  Int32(-32),
		F_Sint64:  Int64(-64),
		F_Enum:    NewDefaults_Color(Defaults_GREEN),
		F_Pinf:    Float32(float32(math.Inf(1))),
		F_Ninf:    Float32(float32(math.Inf(-1))),
		F_Nan:     Float32(1.7),
	}
	SetDefaults(m)
	if !Equal(m, expected) {
		t.Errorf(" got %v\nwant %v", m, expected)
	}
}

func TestNewFromFloat(t *testing.T) {
	var err float64
	for f, s := range testTable {
		d := NewFromFloat(f)
		if d.String() != s {
			err++
			// t.Errorf("expected %s, got %s (%d, %d)",
			// 	s, d.String(), d.compact, d.scale)
		}
	}

	// Some margin of error is acceptable when converting from
	// a float. On a table of roughly 9,000 entries an acceptable
	// margin of error is around 450.
	// Currently, using Gaussian/banker's rounding our margin
	// of error is roughly 215 per 9,000 entries, for a rate of
	// around 2.3%.
	if err >= 0.05*float64(len(testTable)) {
		t.Errorf("expected error rate to be < 0.05%% of table, got %.f", err)
	}

	shouldPanicOn := []float64{
		math.NaN(),
		math.Inf(1),
		math.Inf(-1),
	}

	for _, n := range shouldPanicOn {
		var d *Decimal
		if !didPanic(func() { d = NewFromFloat(n) }) {
			t.Fatalf("expected panic when creating a Decimal from %v, got %v instead", n, d.String())
		}
	}
}

func (b *Button) Initialize() {
	b.Foundation.Initialize()

	b.DrawOp = draw.Over

	b.Label = NewLabel(b.Size, LabelConfig{
		Text:     "",
		FontSize: 12,
		Color:    color.Black,
	})
	b.AddBlock(&b.Label.Block)

	b.Clickers = map[Clicker]bool{}
	b.AddClicker = make(chan Clicker, 1)
	b.RemoveClicker = make(chan Clicker, 1)

	var cs geom.Coord
	cs.X, cs.Y = b.Bounds().Size()
	sh := uik.SizeHint{
		MinSize:       cs,
		PreferredSize: cs,
		MaxSize:       geom.Coord{math.Inf(1), math.Inf(1)},
	}
	b.SetSizeHint(sh)

	b.setConfig = make(chan ButtonConfig, 1)
	b.getConfig = make(chan ButtonConfig, 1)
}

// generateValidatedLengthExample generates a random size array of examples based on what's given.
func (eg *exampleGenerator) generateValidatedLengthExample() interface{} {
	minlength, maxlength := math.Inf(1), math.Inf(-1)
	for _, v := range eg.a.Validations {
		switch actual := v.(type) {
		case *dslengine.MinLengthValidationDefinition:
			minlength = math.Min(minlength, float64(actual.MinLength))
			maxlength = math.Max(maxlength, float64(actual.MinLength))
		case *dslengine.MaxLengthValidationDefinition:
			minlength = math.Min(minlength, float64(actual.MaxLength))
			maxlength = math.Max(maxlength, float64(actual.MaxLength))
		}
	}
	count := 0
	if math.IsInf(minlength, 1) {
		count = int(maxlength) - (eg.r.Int() % 3)
	} else if math.IsInf(maxlength, -1) {
		count = int(minlength) + (eg.r.Int() % 3)
	} else if minlength < maxlength {
		count = int(minlength) + (eg.r.Int() % int(maxlength-minlength))
	} else if minlength == maxlength {
		count = int(minlength)
	} else {
		panic("Validation: MinLength > MaxLength")
	}
	if !eg.a.Type.IsArray() {
		return eg.r.faker.Characters(count)
	}
	res := make([]interface{}, count)
	for i := 0; i < count; i++ {
		res[i] = eg.a.Type.ToArray().ElemType.GenerateExample(eg.r)
	}
	return res
}

// Merge merges the data of two Stats objects.
func (s Stats) Merge(t Stats) Stats {
	if s.count == 0 {
		s.max = math.Inf(-1)
		s.min = math.Inf(+1)
	}

	delta := t.mean - s.mean
	newcount := t.count + s.count

	// max & min
	s.max = math.Max(s.max, t.max)
	s.min = math.Min(s.min, t.min)

	// mean
	s.mean += delta * (t.count / newcount)

	// sum of squares
	s.sumsq += t.sumsq
	s.sumsq += delta * delta * (t.count * s.count / newcount)

	// count
	s.count = newcount

	return s
}

func TestRTT_getDatacenterDistance(t *testing.T) {
	s := newMockServer()

	// The serfer's own DC is always 0 ms away.
	if dist, err := getDatacenterDistance(s, "dc0"); err != nil || dist != 0.0 {
		t.Fatalf("bad: %v err: %v", dist, err)
	}

	// Check a DC with no coordinates, which should give positive infinity.
	if dist, err := getDatacenterDistance(s, "dcX"); err != nil || dist != math.Inf(1.0) {
		t.Fatalf("bad: %v err: %v", dist, err)
	}

	// Similar for a totally unknown DC.
	if dist, err := getDatacenterDistance(s, "acdc"); err != nil || dist != math.Inf(1.0) {
		t.Fatalf("bad: %v err: %v", dist, err)
	}

	// Check the trivial median case (just one node).
	if dist, err := getDatacenterDistance(s, "dc2"); err != nil || dist != 0.002 {
		t.Fatalf("bad: %v err: %v", dist, err)
	}

	// Check the more interesting median case, note that there's a mystery
	// node4 in there that should be excluded to make the distances sort
	// like this:
	//
	// [0] node3 (0.005), [1] node1 (0.007), [2] node2 (0.008)
	//
	// So the median should be at index 3 / 2 = 1 -> 0.007.
	if dist, err := getDatacenterDistance(s, "dc1"); err != nil || dist != 0.007 {
		t.Fatalf("bad: %v err: %v", dist, err)
	}
}

func TestMinimalSurface(t *testing.T) {
	for _, size := range [][2]int{
		{20, 30},
		{30, 30},
		{50, 40},
	} {
		f := NewMinimalSurface(size[0], size[1])
		x0 := f.InitX()
		grad := make([]float64, len(x0))
		f.Grad(grad, x0)
		fdGrad := fd.Gradient(nil, f.Func, x0, &fd.Settings{Formula: fd.Central})

		// Test that the numerical and analytical gradients agree.
		dist := floats.Distance(grad, fdGrad, math.Inf(1))
		if dist > 1e-9 {
			t.Errorf("grid %v x %v: numerical and analytical gradient do not match. |fdGrad - grad|_∞ = %v",
				size[0], size[1], dist)
		}

		// Test that the gradient at the minimum is small enough.
		// In some sense this test is not completely correct because ExactX
		// returns the exact solution to the continuous problem projected on the
		// grid, not the exact solution to the discrete problem which we are
		// solving. This is the reason why a relatively loose tolerance 1e-4
		// must be used.
		xSol := f.ExactX()
		f.Grad(grad, xSol)
		norm := floats.Norm(grad, math.Inf(1))
		if norm > 1e-4 {
			t.Errorf("grid %v x %v: gradient at the minimum not small enough. |grad|_∞ = %v",
				size[0], size[1], norm)
		}
	}
}

func TestScrubValues(t *testing.T) {
	dummy := Converter{
		tracker: new(tracker),
	}

	epoch := time.Unix(0, 0)
	simple := []tsm1.Value{tsm1.NewValue(epoch, 1.0)}

	for _, tt := range []struct {
		input, expected []tsm1.Value
	}{
		{
			input:    simple,
			expected: simple,
		}, {
			input:    []tsm1.Value{simple[0], tsm1.NewValue(epoch, math.NaN())},
			expected: simple,
		}, {
			input:    []tsm1.Value{simple[0], tsm1.NewValue(epoch, math.Inf(-1))},
			expected: simple,
		}, {
			input:    []tsm1.Value{simple[0], tsm1.NewValue(epoch, math.Inf(1)), tsm1.NewValue(epoch, math.NaN())},
			expected: simple,
		},
	} {
		out := dummy.scrubValues(tt.input)
		if !reflect.DeepEqual(out, tt.expected) {
			t.Errorf("Failed to scrub '%s': Got '%s', Expected '%s'", pretty(tt.input), pretty(out), pretty(tt.expected))
		}
	}
}

// Normalize Returns all the values of the given matrix normalized, the formula
// applied to all the elements is: (Xn - Avg) / (max - min) If all the elements
// in the slice have the same values, or the slice is empty, the slice can't be
// normalized, then returns false in the valid parameter
func Normalize(values []float64) (norm []float64, valid bool) {
	avg := 0.0
	max := math.Inf(-1)
	min := math.Inf(1)
	math.Inf(1)
	for _, val := range values {
		avg += val
		if val < min {
			min = val
		}
		if val > max {
			max = val
		}
	}

	if max == min || len(values) == 0 {
		valid = false
		return
	}

	valid = true
	avg /= float64(len(values))
	for _, val := range values {
		norm = append(norm, (val-avg)/(max-min))
	}

	return
}

func TestScaledUpHalfKStandardWeibullProb(t *testing.T) {
	pts := []univariateProbPoint{
		univariateProbPoint{
			loc:     0,
			prob:    math.Inf(1),
			cumProb: 0,
			logProb: math.Inf(1),
		},
		univariateProbPoint{
			loc:     -1,
			prob:    0,
			cumProb: 0,
			logProb: 0,
		},
		univariateProbPoint{
			loc:     1,
			prob:    0.180436508682207,
			cumProb: 0.558022622759326,
			logProb: -1.712376315541750,
		},
		univariateProbPoint{
			loc:     20,
			prob:    0.002369136850928,
			cumProb: 0.974047406098605,
			logProb: -6.045229588092130,
		},
	}
	testDistributionProbs(t, Weibull{K: 0.5, Lambda: 1.5}, "0.5K 1.5λ Weibull", pts)
}