private void FlushCurrentCluster()
{
if (_currentClusterDirty)
{
_clusterWriter.WriteCluster(_currentCluster, _currentClusterBuffer, 0, _currentClusterBuffer.Length);
_currentClusterDirty = false;
}
}
/// <summary>
/// Clusters the data but does not store the results, instead immediately writes the data to the stream writer provided.
/// </summary>
/// <param name="data"></param>
/// <param name="writer"></param>
public void ClusterAndProcess(List <T> data, IClusterWriter <U> writer)
{
/*
* This clustering algorithm first sorts the list of input UMC's by mass. It then iterates
* through this list partitioning the data into blocks of UMC's based on a mass tolerance.
* When it finds gaps larger or equal to the mass (ppm) tolerance specified by the user,
* it will process the data before the gap (a block) until the current index of the features in question.
*/
// Make sure we have data to cluster first.
if (data == null)
{
throw new NullReferenceException("The input feature data list was null. Cannot process this data.");
}
// Make sure there is no null UMC data in the input list.
var nullIndex = data.FindIndex(delegate(T x) { return(x == null); });
if (nullIndex > 0)
{
throw new NullReferenceException("The feature at index " + nullIndex + " was null. Cannot process this data.");
}
OnNotify("Sorting cluster mass list");
// The first thing we do is to sort the features based on mass since we know that has the least variability in the data across runs.
data.Sort(m_massComparer);
// Now partition the data based on mass ranges and the parameter values.
var massTolerance = Parameters.Tolerances.Mass;
// This is the index of first feature of a given mass partition.
var startUMCIndex = 0;
var totalFeatures = data.Count;
OnNotify("Detecting mass partitions");
var tenPercent = Convert.ToInt32(totalFeatures * .1);
var counter = 0;
var percent = 0;
var clusterId = 0;
for (var i = 0; i < totalFeatures - 1; i++)
{
if (counter > tenPercent)
{
counter = 0;
percent += 10;
OnNotify(string.Format("Clustering Completed...{0}%", percent));
}
counter++;
// Here we compute the ppm mass difference between consecutive features (based on mass).
// This will determine if we cluster a block of data or not.
var umcX = data[i];
var umcY = data[i + 1];
var ppm = Math.Abs(FeatureLight.ComputeMassPPMDifference(umcX.MassMonoisotopicAligned, umcY.MassMonoisotopicAligned));
// If the difference is greater than the tolerance then we cluster
// - we dont check the sign of the ppm because the data should be sorted based on mass.
if (ppm > massTolerance)
{
// If start UMC Index is equal to one, then that means the feature at startUMCIndex
// could not find any other features near it within the mass tolerance specified.
if (startUMCIndex == i)
{
var cluster = new U();
cluster.AmbiguityScore = m_maxDistance;
umcX.SetParentFeature(cluster);
cluster.AddChildFeature(umcX);
cluster.CalculateStatistics(Parameters.CentroidRepresentation);
cluster.Id = clusterId++;
writer.WriteCluster(cluster);
}
else
{
// Otherwise we have more than one feature to to consider.
var distances = CalculatePairWiseDistances(startUMCIndex, i, data);
var localClusters = CreateSingletonClusters(data, startUMCIndex, i);
var blockClusters = LinkFeatures(distances, localClusters);
CalculateAmbiguityScore(blockClusters);
foreach (var cluster in localClusters.Values)
{
cluster.Id = clusterId++;
CalculateStatistics(cluster);
writer.WriteCluster(cluster);
}
}
startUMCIndex = i + 1;
}
}
// Make sure that we cluster what is left over.
if (startUMCIndex < totalFeatures)
{
OnNotify(string.Format("Clustering last partition...{0}%", percent));
var distances = CalculatePairWiseDistances(startUMCIndex, totalFeatures - 1, data);
var localClusters = CreateSingletonClusters(data, startUMCIndex, totalFeatures - 1);
var blockClusters = LinkFeatures(distances, localClusters);
CalculateAmbiguityScore(blockClusters);
if (localClusters.Count < 2)
{
foreach (var cluster in localClusters.Values)
{
cluster.Id = clusterId++;
CalculateStatistics(cluster);
writer.WriteCluster(cluster);
}
}
else
{
foreach (var cluster in blockClusters)
{
cluster.Id = clusterId++;
CalculateStatistics(cluster);
writer.WriteCluster(cluster);
}
}
}
// OnNotify("Generating cluster statistics");
}