Class FloatDistribution

Namespace

OptunaSharp.Distributions

Assembly

OptunaSharp.dll

A distribution on floats. Port of optuna/distributions.py: FloatDistribution.

This object is instantiated by SuggestFloat(string, double, double, double?, bool), and passed to samplers in general.

public class FloatDistribution : BaseDistribution, IEquatable<BaseDistribution>

Inheritance

object

BaseDistribution

FloatDistribution

Implements

IEquatable<BaseDistribution>

Derived

DiscreteUniformDistribution

LogUniformDistribution

UniformDistribution

Inherited Members

BaseDistribution.ToExternalRepr(double)

BaseDistribution.Equals(object)

BaseDistribution.ToString()

object.Equals(object, object)

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

Remarks

When step is not null, if the range [low, high] is not divisible by step, high will be replaced with the maximum of k * step + low < high, where k is an integer.

Tiny magnitudes round differently from Optuna (a kept difference). Optuna does this arithmetic with Python's Decimal (28 significant digits, unbounded exponent); OptunaSharp uses .NET decimal, which keeps 28 digits after the point and reads anything below 1e-28 as zero. The owner chose on 2026-11-12 to keep it (QA-3 / QA-62 of the parity QA journals). Ordinary ranges are unaffected; three kinds of input differ:

• a bound or step whose shortest repr needs more than 28 decimal places (a 17-digit value below about 1e-11) can leave high one ULP from Optuna's, and BruteForce's grid one point longer (FloatDistribution(4.1271208719049127e-14, 4.085121318968367e-13, step: 1e-15): 4.082712087190491e-13, Optuna 4.0827120871904913e-13);
• a bound below 1e-28 counts as zero, so (5e-324, 0.30000000000000004, step: 1.0) ends at 0.0, below its low (Optuna: 5e-324);
• a range whose step count needs more than 28 digits ((0, 1e20, step: 1e-09)) is accepted, where Optuna raises decimal.InvalidOperation; a NaN or infinite bound or step throws FormatException, where Optuna gives high=nan or InvalidOperation.

Constructors

FloatDistribution(double, double, bool, double?)

Creates a numeric distribution and normalizes a stepped upper endpoint.

public FloatDistribution(double low, double high, bool log = false, double? step = null)

Parameters

low double

Inclusive lower endpoint of the parameter range.

high double

Inclusive upper endpoint; a stepped range may lower it to the last reachable grid point.

log bool

Whether to sample in the logarithmic domain, which requires a strictly positive lower endpoint.

step double?

Positive float grid spacing, or null for a continuous range; logarithmic ranges require null.

Exceptions

ValueError

The lower endpoint exceeds the upper endpoint, logarithmic bounds are nonpositive, or the step is invalid for the selected scale.

FormatException

A stepped bound or step is NaN or infinite and cannot be parsed as a decimal.

DivideByZeroException

A positive step rounds to zero in decimal arithmetic while aligning the upper endpoint.

Properties

High

Upper endpoint of the range of the distribution. High is included in the range.

public double High { get; }

Property Value

double

Log

If true, this distribution is in log-scaled domain.

public bool Log { get; }

Property Value

bool

Low

Lower endpoint of the range of the distribution. Low is included in the range.

public double Low { get; }

Property Value

double

Step

A discretization step. null when the distribution is continuous.

public double? Step { get; }

Property Value

double?

Methods

Equals(BaseDistribution?)

Compares distribution types and their defining attributes.

public override bool Equals(BaseDistribution? other)

Parameters

other BaseDistribution

Value to compare with this instance; null compares unequal.

Returns

bool

True when the distributions have compatible types and equal attributes.

GetHashCode()

Computes a hash from the distribution type and defining attributes.

public override int GetHashCode()

Returns

int

Hash code consistent with distribution equality.

Single()

Tests whether this distribution admits only one candidate.

public override bool Single()

Returns

bool

True when sampling can be replaced by the sole allowed value.

Exceptions

FormatException

A stepped bound or step is NaN or infinite and cannot be parsed as a decimal.

ToInternalRepr(object?)

Converts an external value into the numeric representation used by samplers.

public override double ToInternalRepr(object? paramValueInExternalRepr)

Parameters

paramValueInExternalRepr object

External parameter value to convert to the distribution's numeric representation.

Returns

double

The converted numeric value; range membership is checked separately by Contains.

Exceptions

ValueError

The value cannot be converted to a number, is NaN, or is nonpositive for a logarithmic distribution.