def func(alpha, nClusters=nClusters):
alpha1, alpha2 = alpha[:self.nVars], alpha[self.nVars:]
alpha1 = alpha1.reshape((self.nVars, 1))
alpha2 = alpha2.reshape(self.nVars, 1)
alpha = numpy.append(alpha1, alpha2, axis=1)
smpl = numpy.dot(self.Z, alpha)
smpl = orange.ExampleTable(smpl)
km1 = orngClustering.KMeans(smpl, centroids=nClusters)
km2 = orngClustering.KMeans(smpl, centroids=nClusters)
score = (self.totalSize - nClusters -
1) * (km1.score - km2.score) / (km2.score)
return -score
def runOptimization(self):
if self.optimizationTo > len(set(self.data)):
self.error("Not enough unique data instances (%d) for given number of clusters (%d)." % \
(len(set(self.data)), self.optimizationTo))
return
random.seed(0)
try:
self.progressBarInit()
Ks = list(range(self.optimizationFrom, self.optimizationTo + 1))
self.optimizationRun =[(k, orngClustering.KMeans(
self.data,
centroids = k,
minscorechange=0,
nstart = self.restarts,
initialization = self.initializations[self.initializationType][1],
distance = self.distanceMeasures[self.distanceMeasure][1],
scoring = self.scoringMethods[self.scoring][1],
inner_callback = lambda val: self.progressBarSet(min(self.progressEstimate(val)/len(Ks) + k * 100.0 / len(Ks), 100.0))
)) for k in Ks]
self.progressBarFinished()
self.bestRun = (min if getattr(self.scoringMethods[self.scoring][1], "minimize", False) else max)(self.optimizationRun, key=lambda k_run: k_run[1].score)
self.showResults()
self.sendData()
except Exception as ex:
self.error(0, "An error occured while running optimization. Reason: " + str(ex))
raise
def runOptimization(self):
if self.optimizationTo > len(set(self.data)):
self.error("Not enough unique data instances (%d) for given number of clusters (%d)." % \
(len(set(self.data)), self.optimizationTo))
return
random.seed(0)
try:
self.progressBarInit()
Ks = range(self.optimizationFrom, self.optimizationTo + 1)
outer_callback_count = len(Ks) * self.restarts
outer_callback_state = {"restart": 0}
optimizationRun = []
for k in Ks:
def outer_progress(km):
outer_callback_state["restart"] += 1
self.progressBarSet(100.0 *
outer_callback_state["restart"] /
outer_callback_count)
def inner_progress(km):
estimate = self.progressEstimate(km)
self.progressBarSet(min(estimate / outer_callback_count + \
outer_callback_state["restart"] * \
100.0 / outer_callback_count,
100.0))
kmeans = orngClustering.KMeans(
self.data,
centroids=k,
minscorechange=0,
nstart=self.restarts,
initialization=self.initializations[
self.initializationType][1],
distance=self.distanceMeasures[self.distanceMeasure][1],
scoring=self.scoringMethods[self.scoring][1],
outer_callback=outer_progress,
inner_callback=inner_progress)
optimizationRun.append((k, kmeans))
if self.restarts == 1:
outer_progress(None)
self.optimizationRun = optimizationRun
self.progressBarFinished()
self.bestRun = (min if getattr(
self.scoringMethods[self.scoring][1], "minimize", False) else
max)(self.optimizationRun,
key=lambda (k, run): run.score)
self.showResults()
self.sendData()
except Exception, ex:
self.error(
0, "An error occured while running optimization. Reason: " +
str(ex))
raise
def func(alpha, nClusters=nClusters):
alpha1, alpha2 = alpha[:self.nVars], alpha[self.nVars:]
alpha1 = alpha1.reshape((self.nVars, 1))
alpha2 = alpha2.reshape(self.nVars, 1)
alpha = numpy.append(alpha1, alpha2, axis=1)
smpl = numpy.dot(self.Z, alpha)
smpl = orange.ExampleTable(smpl)
km = orngClustering.KMeans(smpl, centroids=nClusters)
score = orngClustering.score_silhouette(km)
return -score
def cluster(self):
if self.K > len(set(self.data)):
self.error("Not enough unique data instances (%d) for given number of clusters (%d)." % \
(len(set(self.data)), self.K))
return
random.seed(0)
self.km = orngClustering.KMeans(
centroids=self.K,
minscorechange=0,
nstart=self.restarts,
initialization=self.initializations[self.initializationType][1],
distance=self.distanceMeasures[self.distanceMeasure][1],
scoring=self.scoringMethods[self.scoring][1],
inner_callback=self.clusterCallback,
)
self.progressBarInit()
self.km(self.data)
self.sendData()
self.progressBarFinished()
import orange
import orngClustering
import random
data_names = ["iris.tab", "housing.tab", "vehicle.tab"]
data_sets = [orange.ExampleTable(name) for name in data_names]
print "%10s %3s %3s %3s" % ("", "Rnd", "Div", "HC")
for data, name in zip(data_sets, data_names):
random.seed(42)
km_random = orngClustering.KMeans(data, centroids=3)
km_diversity = orngClustering.KMeans(data, centroids=3, \
initialization=orngClustering.kmeans_init_diversity)
km_hc = orngClustering.KMeans(data, centroids=3, \
initialization=orngClustering.KMeans_init_hierarchicalClustering(n=100))
print "%10s %3d %3d %3d" % (name, km_random.iteration,
km_diversity.iteration, km_hc.iteration)
import orange
import orngClustering
import random
random.seed(42)
data = orange.ExampleTable("iris")
km = orngClustering.KMeans(data, 3)
print km.clusters[-10:]
import orange
import orngClustering
import random
data = orange.ExampleTable("iris")
# data = orange.ExampleTable("lung-cancer")
bestscore = 0
for k in range(2, 10):
random.seed(42)
km = orngClustering.KMeans(
data,
k,
initialization=orngClustering.KMeans_init_hierarchicalClustering(n=50),
nstart=10)
score = orngClustering.score_silhouette(km)
print "%d: %.3f" % (k, score)
if score > bestscore:
best_km = km
bestscore = score
orngClustering.plot_silhouette(best_km, filename='tmp.png')
import orange
import orngClustering
data = orange.ExampleTable("voting")
# data = orange.ExampleTable("iris")
for k in range(2, 5):
km = orngClustering.KMeans(
data, k, initialization=orngClustering.kmeans_init_diversity)
score = orngClustering.score_silhouette(km)
print k, score
km = orngClustering.KMeans(data,
3,
initialization=orngClustering.kmeans_init_diversity)
orngClustering.plot_silhouette(km, "kmeans-silhouette.png")
import orange
import orngClustering
import random
random.seed(42)
def callback(km):
print "Iteration: %d, changes: %d, score: %.4f" % (km.iteration,
km.nchanges, km.score)
data = orange.ExampleTable("iris")
km = orngClustering.KMeans(data, 3, minscorechange=0, inner_callback=callback)
xc = [float(d[attx]) for d in km.centroids]
yc = [float(d[atty]) for d in km.centroids]
pylab.scatter(xc, yc, marker="x", c="k", s=200)
pylab.xlabel(attx)
pylab.ylabel(atty)
if title:
pylab.title(title)
pylab.savefig("%s-%03d.png" % (filename, km.iteration))
pylab.close()
def in_callback(km):
print "Iteration: %d, changes: %d, score: %8.6f" % (km.iteration,
km.nchanges, km.score)
plot_scatter(data,
km,
"petal width",
"petal length",
title="Iteration %d" % km.iteration)
data = orange.ExampleTable("iris")
random.seed(42)
km = orngClustering.KMeans(data,
3,
minscorechange=0,
maxiters=10,
inner_callback=in_callback)