/// <summary>
/// Estimates an Empirical Hazards distribution considering event times and the outcome of the
/// observed sample at the time of event, plus additional parameters for the hazard estimation.
/// </summary>
///
/// <param name="time">The time of occurrence for the event.</param>
/// <param name="outcome">The outcome at the time of event (failure or censored).</param>
/// <param name="weights">The weights associated with each event.</param>
/// <param name="hazard">The hazard estimator to use. Default is <see cref="HazardEstimator.BreslowNelsonAalen"/>.</param>
/// <param name="survival">The survival estimator to use. Default is <see cref="SurvivalEstimator.FlemingHarrington"/>.</param>
/// <param name="ties">The method for handling event ties. Default is <see cref="HazardTiesMethod.Efron"/>.</param>
///
/// <returns>The <see cref="EmpiricalHazardDistribution"/> estimated from the given data.</returns>
///
public static EmpiricalHazardDistribution Estimate(double[] time, SurvivalOutcome[] outcome,
double[] weights, SurvivalEstimator survival, HazardEstimator hazard, HazardTiesMethod ties)
{
var dist = new EmpiricalHazardDistribution(survival);
dist.Fit(time, weights, new EmpiricalHazardOptions
{
Outcome = outcome,
Estimator = hazard,
Ties = ties
});
return(dist);
}
void init(double[] times, double[] hazards, SurvivalEstimator estimator)
{
if (times == null)
{
times = new double[0];
}
if (hazards == null)
{
hazards = new double[0];
}
times = (double[])times.Clone();
hazards = (double[])hazards.Clone();
Array.Sort(times, hazards);
for (int i = 0; i < times.Length - 1; i++)
{
if (times[i] > times[i + 1])
{
throw new Exception();
}
}
if (times.Length > 0)
{
this.range = new DoubleRange(0, times[times.Length - 1]);
this.survivals = new double[times.Length];
this.survivals[0] = 1;
for (int i = 1; i < survivals.Length; i++)
{
survivals[i] = survivals[i - 1] * (1.0 - hazards[i - 1]);
}
}
this.times = times;
this.hazards = hazards;
this.estimator = estimator;
this.mean = null;
this.variance = null;
}
/// <summary>
/// Initializes a new instance of the <see cref="EmpiricalHazardDistribution"/> class.
/// </summary>
///
/// <param name="times">The time steps.</param>
/// <param name="lambdas">The hazard rates at the time steps.</param>
/// <param name="estimator">The survival function estimator to be used. Default is
/// <see cref="SurvivalEstimator.FlemingHarrington"/></param>
///
public EmpiricalHazardDistribution(double[] times, double[] lambdas, SurvivalEstimator estimator)
{
if (times == null)
{
throw new ArgumentNullException("time");
}
if (lambdas == null)
{
throw new ArgumentNullException("values");
}
if (times.Length != lambdas.Length)
{
throw new DimensionMismatchException("time",
"The time steps and value vectors must have the same length.");
}
init(times, lambdas, estimator);
}
/// <summary>
/// Estimates an Empirical Hazards distribution considering event times and the outcome of the
/// observed sample at the time of event, plus additional parameters for the hazard estimation.
/// </summary>
///
/// <param name="time">The time of occurrence for the event.</param>
/// <param name="outcome">The outcome at the time of event (failure or censored).</param>
/// <param name="hazard">The hazard estimator to use. Default is <see cref="HazardEstimator.BreslowNelsonAalen"/>.</param>
/// <param name="survival">The survival estimator to use. Default is <see cref="SurvivalEstimator.FlemingHarrington"/>.</param>
///
/// <returns>The <see cref="EmpiricalHazardDistribution"/> estimated from the given data.</returns>
///
public static EmpiricalHazardDistribution Estimate(double[] time, int[] outcome,
SurvivalEstimator survival, HazardEstimator hazard)
{
return(Estimate(time, outcome, survival, hazard, EmpiricalHazardOptions.DefaultTies));
}
/// <summary>
/// Fits the underlying distribution to a given set of observations.
/// </summary>
///
/// <param name="observations">The array of observations to fit the model against. The array
/// elements can be either of type double (for univariate data) or
/// type double[] (for multivariate data).</param>
/// <param name="weights">The weight vector containing the weight for each of the samples.</param>
/// <param name="options">Optional arguments which may be used during fitting, such
/// as regularization constants and additional parameters.</param>
///
public void Fit(double[] observations, double[] weights, SurvivalOptions options)
{
SurvivalOutcome[] outcome = null;
double[] output = weights;
var estimator = HazardEstimator.BreslowNelsonAalen;
var ties = HazardTiesMethod.Breslow;
if (options != null)
{
outcome = options.Outcome;
var hazardOps = options as EmpiricalHazardOptions;
if (hazardOps != null)
{
estimator = hazardOps.Estimator;
ties = hazardOps.Ties;
}
}
if (outcome == null)
{
outcome = Vector.Create(observations.Length, SurvivalOutcome.Failed);
}
if (output == null)
{
output = Vector.Create(observations.Length, 1.0);
}
if (estimator == HazardEstimator.KaplanMeier)
{
double[] times = observations.Distinct();
double[] hazards = new double[times.Length];
hazards[hazards.Length - 1] = 1;
Array.Sort(times);
// Compute an estimate of the hazard function using
// the Kaplan-Meier estimator for the survival function
double r = observations.Length;
double d = 0; // Number of deaths
double c = 0; // Number of censor
for (int j = 0; j < times.Length; j++)
{
double t = times[j];
// Survivals until time t
r = r - d - c;
d = 0;
c = 0;
for (int i = 0; i < observations.Length; i++)
{
if (observations[i] == t)
{
if (outcome[i] == SurvivalOutcome.Failed)
{
d++; // Failure at time t
}
else
{
c++; // Censored at time t
}
}
}
hazards[j] = d / r;
}
this.init(times, hazards, this.estimator);
return;
}
else if (estimator == HazardEstimator.BreslowNelsonAalen)
{
double[] survivals = new double[observations.Length];
Sort(ref observations, ref outcome, ref output);
// Compute an estimate of the cumulative Hazard
// function using the Breslow-Nelson-Aalen estimator
double sum = 0;
double d = 0;
for (int i = 0; i < observations.Length; i++)
{
survivals[i] = 1;
double t = observations[i];
double v = output[i];
var o = outcome[i];
sum += v;
if (o == SurvivalOutcome.Censored)
{
d = 0;
continue;
}
if (i > 0 && t != observations[i - 1])
{
d = 0;
}
// Count the number of deaths at t
if (o == SurvivalOutcome.Failed)
{
d++; // Deaths at time t
}
if (i < observations.Length - 1 && t == observations[i + 1] && outcome[i + 1] == SurvivalOutcome.Failed)
{
continue;
}
if (ties == HazardTiesMethod.Breslow)
{
survivals[i] = Math.Exp(-d / sum);
}
else
{
if (d == 1)
{
survivals[i] = Math.Pow(1 - v / sum, 1 / v);
}
else
{
survivals[i] = Math.Exp(-d / sum);
}
}
}
// Transform to hazards
double[] hazards = new double[observations.Length];
for (int i = 0; i < hazards.Length; i++)
{
hazards[i] = -Math.Log(survivals[i]);
}
Array.Sort(observations, hazards);
this.init(observations, hazards, this.estimator);
return;
}
throw new ArgumentException();
}
/// <summary>
/// Initializes a new instance of the <see cref="EmpiricalHazardDistribution"/> class.
/// </summary>
///
/// <param name="estimator">The survival function estimator to be used. Default is
/// <see cref="SurvivalEstimator.FlemingHarrington"/></param>
///
public EmpiricalHazardDistribution(SurvivalEstimator estimator)
{
init(null, null, estimator);
}
/// <summary>
/// Estimates an Empirical Hazards distribution considering event times and the outcome of the
/// observed sample at the time of event, plus additional parameters for the hazard estimation.
/// </summary>
///
/// <param name="time">The time of occurrence for the event.</param>
/// <param name="outcome">The outcome at the time of event (failure or censored).</param>
/// <param name="hazard">The hazard estimator to use. Default is <see cref="HazardEstimator.BreslowNelsonAalen"/>.</param>
/// <param name="survival">The survival estimator to use. Default is <see cref="SurvivalEstimator.FlemingHarrington"/>.</param>
/// <param name="ties">The method for handling event ties. Default is <see cref="HazardTiesMethod.Efron"/>.</param>
///
/// <returns>The <see cref="EmpiricalHazardDistribution"/> estimated from the given data.</returns>
///
public static EmpiricalHazardDistribution Estimate(double[] time, SurvivalOutcome[] outcome,
SurvivalEstimator survival, HazardEstimator hazard, HazardTiesMethod ties)
{
return(Estimate(time, outcome, null, survival, hazard, ties));
}
/// <summary>
/// Estimates an Empirical Hazards distribution considering event times and the outcome of the
/// observed sample at the time of event, plus additional parameters for the hazard estimation.
/// </summary>
///
/// <param name="time">The time of occurrence for the event.</param>
/// <param name="outcome">The outcome at the time of event (failure or censored).</param>
/// <param name="weights">The weights associated with each event.</param>
/// <param name="hazard">The hazard estimator to use. Default is <see cref="HazardEstimator.BreslowNelsonAalen"/>.</param>
/// <param name="survival">The survival estimator to use. Default is <see cref="SurvivalEstimator.FlemingHarrington"/>.</param>
/// <param name="ties">The method for handling event ties. Default is <see cref="HazardTiesMethod.Efron"/>.</param>
///
/// <returns>The <see cref="EmpiricalHazardDistribution"/> estimated from the given data.</returns>
///
public static EmpiricalHazardDistribution Estimate(double[] time, int[] outcome,
double[] weights, SurvivalEstimator survival, HazardEstimator hazard, HazardTiesMethod ties)
{
return(Estimate(time, outcome.To <SurvivalOutcome[]>(), weights, survival, hazard, ties));
}
/// <summary>
/// Estimates an Empirical Hazards distribution considering event times and the outcome of the
/// observed sample at the time of event, plus additional parameters for the hazard estimation.
/// </summary>
///
/// <param name="time">The time of occurrence for the event.</param>
/// <param name="weights">The weights associated with each event.</param>
/// <param name="hazard">The hazard estimator to use. Default is <see cref="HazardEstimator.BreslowNelsonAalen"/>.</param>
/// <param name="survival">The survival estimator to use. Default is <see cref="SurvivalEstimator.FlemingHarrington"/>.</param>
/// <param name="ties">The method for handling event ties. Default is <see cref="HazardTiesMethod.Efron"/>.</param>
///
/// <returns>The <see cref="EmpiricalHazardDistribution"/> estimated from the given data.</returns>
///
public static EmpiricalHazardDistribution Estimate(double[] time, double[] weights = null,
SurvivalEstimator survival = EmpiricalHazardOptions.DefaultSurvival,
HazardEstimator hazard = EmpiricalHazardOptions.DefaultEstimator, HazardTiesMethod ties = EmpiricalHazardOptions.DefaultTies)
{
return(Estimate(time, (SurvivalOutcome[])null, weights, survival, hazard, ties));
}
