def getBestKValue(zipcodes, maxClusters):
if maxClusters == 1:
return maxClusters
else:
gaps, s_k, K = gap.gap_statistic(zipcodes,
refs=None,
B=10,
K=range(2, maxClusters),
N_init=10)
if len(gaps) == 0:
print 1
return 1
bestKValue = gap.find_optimal_k(gaps, s_k, K)
return bestKValue
def main():
file_path = sys.argv[1]
output_path = sys.argv[2]
#Obtain the SOM centroids and cluster them.
#Open the file containing the SOM centroids.
#Add all centroids to the list of data to cluster.
som_centroids = []
if os.path.exists(file_path):
file = open(file_path, 'r')
next_line = file.readline()
while next_line:
som_centroids.append([float(i) for i in next_line.split(",")])
next_line = file.readline()
#Compute the gap statistic and use it to find the best k-value.
gaps, s_k, K = gap.gap_statistic(np.array(som_centroids),
refs=None,
B=10,
K=range(1, len(som_centroids)),
N_init=10)
bestKValue = gap.find_optimal_k(gaps, s_k, K)
#Print message to user.
print("Optimal K is " + str(bestKValue))
#Perform k-means clustering.
kmeans = KMeans(n_clusters=bestKValue,
random_state=0).fit(np.array(som_centroids))
#Print cluster centroids from hierarchical clustering.
file = open(output_path, 'w')
print_centroids(kmeans.cluster_centers_, som_centroids, file)
#Print message to user.
print("Clustering complete.")
def eval(data, refs=None, nrefs=20, ks=range(2, 100), n_init=10):
"""
Compute the Gap statistic for an nxm dataset in data using gapkmean (from standard Python library)
Either give a precomputed set of reference distributions in refs as an (n,m,k) scipy array,
or state the number k of reference distributions in nrefs for automatic generation with a
uniformed distribution within the bounding box of data.
Give the list of k-values for which you want to compute the statistic in ks.
Reference
---------
https://pypi.python.org/pypi/gapkmean/1.0
"""
from gap import gap
gaps, s_k, K = gap.gap_statistic(data,
refs=refs,
B=nrefs,
K=ks,
N_init=n_init)
opt_k = gap.find_optimal_k(gaps, s_k, K)
return opt_k