func newCluster(keys []string, regression float64, hash string) *types.ClusterSummary {
c := types.NewClusterSummary(len(keys), 0)
for i, k := range keys {
c.Keys[i] = k
}
c.StepFit.Regression = regression
c.Hash = hash
return c
}
// GetClusterSummaries returns a summaries for each cluster.
func GetClusterSummaries(observations []kmeans.Clusterable, centroids []kmeans.Centroid, commits []*tiling.Commit) *ClusterSummaries {
ret := &ClusterSummaries{
Clusters: make([]*types.ClusterSummary, len(centroids)),
}
allClusters, _ := kmeans.GetClusters(observations, centroids)
for i, cluster := range allClusters {
// cluster is just an array of the observations for a given cluster.
numSampleTraces := len(cluster)
if numSampleTraces > config.MAX_SAMPLE_TRACES_PER_CLUSTER {
numSampleTraces = config.MAX_SAMPLE_TRACES_PER_CLUSTER
}
stepFit := getStepFit(cluster[0].(*ctrace.ClusterableTrace).Values)
summary := types.NewClusterSummary(len(cluster)-1, numSampleTraces)
summary.ParamSummaries = getParamSummaries(cluster)
summary.StepFit = stepFit
summary.Hash = commits[stepFit.TurningPoint].Hash
summary.Timestamp = commits[stepFit.TurningPoint].CommitTime
for j, o := range cluster {
if j != 0 {
summary.Keys[j-1] = o.(*ctrace.ClusterableTrace).Key
}
}
// First, sort the traces so they are order with the traces closest to the
// centroid first.
sc := []*SortableClusterable{}
for j := 0; j < len(cluster); j++ {
sc = append(sc, &SortableClusterable{
Cluster: cluster[j],
Distance: centroids[i].Distance(cluster[j]),
})
}
// Sort, but leave the centroid, the 0th element, unmoved.
sort.Sort(SortableClusterableSlice(sc[1:]))
for j, clusterable := range sc {
if j == numSampleTraces {
break
}
summary.Traces[j] = traceToFlot(clusterable.Cluster.(*ctrace.ClusterableTrace))
}
ret.Clusters[i] = summary
}
sort.Sort(SortableClusterSummarySlice(ret.Clusters))
return ret
}