def kmeansClustering(cluster_list, k, iterations, shuffle = True):
"""
Compute the k-means clustering of a set of clusters (reads/kmers)
Note: the function may not mutate cluster_list
Input: List of clusters, k number of clusters, iterations,
select initial clusters: randomly or by size?
Output: List of clusters.
"""
kclusters = [] # this list to store k clusters to compare with (non-mutable)
centroids = [] # this list to store the initial k centroids (average stats vectors)
if shuffle:
# shuffle cluster list
random.shuffle(cluster_list)
else:
# sort by size
cluster_list.sort(key = lambda cluster: cluster.getSize(), reverse = True)
# k initial clusters to define initial centroids
for cluster in cluster_list[:k]:
kclusters.append(cluster.copy())
centroids.append(cluster.getAvgAbundance())
for iteration in range(iterations):
clusters = []
# initialize new empty cluster objects at the centroids
for idx in range(k):
cluster = Cluster([])
cluster.avg_abundance_vectors = list(centroids[idx])
clusters.append(cluster)
# for every cluster in cluster_list
for num in range(len(cluster_list)):
best = (float('inf'), -1)
# compare distance to every centroid at kclusters
for idx in range(k):
temp = cluster_list[num].distance(kclusters[idx])
if temp < best[0]:
best = (temp, idx)
# merge cluster to best centroid in list of mutable clusters
clusters[best[1]].mergeClusters(cluster_list[num])
# make a copy of re-computed centroids: kclusters and centroids.
for idx in range(k):
kclusters[idx] = clusters[idx].copy()
centroids[idx] = (clusters[idx].getAvgAbundance())
return kclusters
