public List <MultivariateNegativeBinomial> InitializeNegativeBinomialEmission(List <List <double> > data, int nHiddenStates, List <double> haploidMean)
{
int nDimensions = _nSamples;
var variance = new List <double>(nDimensions);
var tmpDistributions = new List <MultivariateNegativeBinomial>();
for (int dimension = 0; dimension < nDimensions; dimension++)
{
double meanHolder = data.Sum(datapoint => datapoint[dimension]);
haploidMean.Add(meanHolder / data.Count / 2.0);
variance.Add(CanvasCommon.Utilities.Variance(data.Select(x => x[dimension]).ToList()));
}
// remove outliers
double maxThreshold = haploidMean.Max() * nHiddenStates;
RemoveOutliers(data, maxThreshold);
var maxValues = data.Select(x => Convert.ToInt32(x.Max())).ToList().Max();
for (int CN = 0; CN < nHiddenStates; CN++)
{
var tmpMean = Vector <double> .Build.Dense(haploidMean.Select(x => Math.Max(CN, 0.1) * x).ToArray());
// if few hidden states, increase the last CN state by diploid rather than haploid increment
if (nHiddenStates < 5 && CN - 1 == nHiddenStates)
{
tmpMean = Vector <double> .Build.Dense(haploidMean.Select(x => Math.Max(CN, 0.5) * x + x).ToArray());
}
var tmpDistribution = new MultivariateNegativeBinomial(tmpMean.ToList(), variance, maxValues + 10);
tmpDistributions.Add(tmpDistribution);
}
return(tmpDistributions);
}
public List <MultivariateNegativeBinomial> InitializeNegativeBinomialEmission(List <List <double> > data, int nHiddenStates, List <double> haploidMean, List <double> medians = null, List <double> pseudoVariances = null)
{
int nDimensions = _nSamples;
var variance = new List <double>(nDimensions);
var tmpDistributions = new List <MultivariateNegativeBinomial>();
for (int dimension = 0; dimension < nDimensions; dimension++)
{
double median = Math.Max(1d, CanvasCommon.Utilities.Median(data.Select(datapoint => datapoint[dimension])));
if (medians == null)
{
haploidMean.Add(median / 2.0);
variance.Add(CanvasCommon.Utilities.Variance(data.Select(x => x[dimension]).ToList()));
}
else
{
haploidMean.Add(medians[dimension] / 2.0);
variance.Add(pseudoVariances[dimension]);
}
}
// remove outliers
double maxThreshold = haploidMean.Max() * nHiddenStates;
RemoveOutliers(data, maxThreshold);
var maxValues = data.Select(x => Convert.ToInt32(x.Max())).ToList().Max();
for (int CN = 0; CN < nHiddenStates; CN++)
{
var tmpMean = Vector <double> .Build.Dense(haploidMean.Select(x => Math.Max(CN, 0.1) * x).ToArray());
// if few hidden states, increase the last CN state by diploid rather than haploid increment
if (nHiddenStates < 5 && CN - 1 == nHiddenStates)
{
tmpMean = Vector <double> .Build.Dense(haploidMean.Select(x => Math.Max(CN, 0.5) * x + x).ToArray());
}
//var tmpVar = pseudoVariances == null ?
// variance
// : variance.Select(x => (Math.Max(CN, 0.5) / 2d) * (Math.Max(CN, 0.5) / 2d) * x);
var tmpVar = variance;
var tmpDistribution = new MultivariateNegativeBinomial(tmpMean.ToList(), tmpVar.ToList(), maxValues + 10);
tmpDistributions.Add(tmpDistribution);
}
return(tmpDistributions);
}
