def extension(self,q1=None):
from itertools import imap, chain
from gmode_module import Invert, free, hyp_test
if q1 == None: q1 = self.q1
cluster_members = self.cluster_members
sample = self.elems[self.excluded]
self.reclass = deque()
for n, st in enumerate(self.cluster_stats):
self.reclass.append(list())
iS = Invert(st[2]) #, Invert(st[3])
f = free(st[3])
size = len(cluster_members[n])
selected = filter(lambda x: x != None, \
imap(lambda ind, y: hyp_test(size,q1,f,ind,y,st[0],iS), self.excluded, sample))
if len(selected) != 0:
self.reclass[n].extend(selected)
N = set(chain.from_iterable(reclass))
self.report.append("\n Reclassified Excluded Sample Size: "+str(len(N)))
print("Reclassified : ",len(N))
self.report.append("\n Totally Excluded: "+str(len(sample) - len(N)))
def correspondence(self, templates, template_name, q1=None, artifact=None, var=None):
''' Fulchignoni et al. (2000) extension used to give a correspondence to clusters'''
from itertools import imap
from gmode_module import Invert, free, hyp_test
from file_module import pickle, writedict
import cPickle as pkl
if q1 == None: q1 = self.q1
cluster_members = self.cluster_members
cluster_stats = self.cluster_stats
templ = pkl.load(open(templates,'rb'))
interpretation = dict()
for n, stat in enumerate(cluster_stats):
iS = Invert(stat[2][var, :][:, var])
f = free(stat[3][var, :][:, var])
size = len(cluster_members[n])
selected = filter(lambda x: x != None, \
imap(lambda key, y: hyp_test(size, q1, f, key, y[var], stat[0][var], iS), templ.keys(), templ.values()))
interpretation[n+1] = selected
writedict(interpretation,open(pathjoin("TESTS",self.label,'correspondence_q'+str(q1)+'_'+template_name+'.dat'), 'w'))
pickle(interpretation, self.label, "correspondence_q"+str(q1)+'_'+template_name)
def clustering(q1, ulim, mlim, grid, design, data, report):
''' Iterative procedure of clustering'''
N = data.shape[0] # Sample Size
M = data.shape[1] # Variable size
#________________________________Barycenter______________________________________
seed = barycenter_density(data, grid, amax(data, axis=0), amin(data, axis=0), sqrt(mlim) )#nmin=int(2*grid*30/M))
#seed = barycenter_hist(grid, design, data)
if len(seed) > 2:
report.append("Barycenter: "+l_to_s(design[seed])) # map(lambda j: design[j], seed)))
#print("Barycenter: "+l_to_s(map(lambda j: design[j], seed)))
Na = len(seed)
cluster = list(seed)
else:
return [], seed, report, 0.0
# ______________________________CLASSIFICATION__________________________________
i = 0
f_rec = deque()
cluster = seed
index = range(N)
while (i == 0 or cluster[:] != cluster_prior[:]) and i < 30 and Na > 2:
# *c --> cluster statistics
ct, std, S, r2 = stats(data[cluster])
# Replace lower deviations than minimal limit:
if i == 0 and any(S < mlim):
S = mlim
std = sqrt(diagonal(S))
iS = Invert(S)
f = free(r2)
cluster_prior = cluster
#f_rec.append(f)
#f_ = min(f_rec)
# Once upper limit is reached the iteration is haulted.
if ulim < 1e0 and any(std >= ulim):
return cluster, seed, report, Na*f
report.append("Run "+str(i)+" Size: "+str(Na)+" S.D.: "+l_to_s(arround(std, 3))+"\nf: "+str(f)+"\n")
# G hypothesis test:
if i == 0 or Na*f >= 30:
cluster = filter(lambda x: x != None, \
imap(lambda ind, y: hyp_test(Na,q1,f,ind,y,ct,iS), index, data))
else:
return cluster, seed, report, Na*f
Na = len(cluster)
# Discreetly increase std. until the seed start growing by itself.
#if i == 0 and Na <= Na_prior:
# S = S + mlim
# std = sqrt(diagonal(S))
#else:
i += 1
return cluster, seed, report, Na*f