def createClustersAndMatchingMatrices(parts, atoms, numOfClusters, disFactor):
vecs = []
mats = []
for atom in atoms:
closeVectors = []
tmp = []
fig = plt.figure()
i=1
for part in parts:
retVal = getAtomFromFrac(part, atom, simpleDis)
if retVal is None:
continue
frac, dis, vecAtom, bias, bestScale, temporalScale, offset, _= retVal
tmp.append((dis, frac))
amplitude = np.max(vecAtom) - np.min(vecAtom)
if(dis > amplitude*disFactor):
continue
closeVectors.append(frac)
if(i>25):
continue
curr = fig.add_subplot(5,5,i)
i+=1
plt.title('bias: ' + str(bias) + ' scale: ' + str(bestScale) +
' bestTemporalScale: ' + str(temporalScale) + ' dis: ' + str(dis))
curr.plot(part, c='b')
rng = xrange(offset, offset + len(frac))
curr.plot(rng, vecAtom, c='g')
curr.plot(rng, frac, c='r')
#plt.show()
print len(tmp), len(closeVectors)
if(len(closeVectors) < numOfClusters):
tmp = sorted(tmp, key=lambda tup: tup[0])
clusteredVectors = [t[1] for t in tmp[:numOfClusters]]
else:
lengthCounter = {}
for vec in closeVectors:
lengthCounter[len(vec)] = lengthCounter.get(len(vec), []) + [vec]
residum = len(closeVectors)%numOfClusters
clusteredVectors = []
tmpNumOfClusters = numOfClusters
last = len(lengthCounter) - 1
for i, (vecLen, sameLengthVecs) in enumerate(lengthCounter.items()):
#if(vecLen <= len(sameLengthVecs)):
#sameLengthVecs, idx = kmeans2(np.array(sameLengthVecs), max(1, int(float(numOfClusters)*vecLen/len(closeVectors))))
numOfClusterPerLength = int(round(float(numOfClusters)*len(sameLengthVecs)/len(closeVectors)))
numOfClusterPerLength = min(numOfClusterPerLength, tmpNumOfClusters) if i!=last else tmpNumOfClusters
tmpNumOfClusters -= numOfClusterPerLength
if(numOfClusterPerLength == 0):
continue
sameLengthVecs, idx = kmeans(np.array(sameLengthVecs), numOfClusterPerLength)
for v in sameLengthVecs:
clusteredVectors.append(pe.toList(v))
#clusteredVectors = clusteredVectors + sameLengthVecs
vecs.append(pe.toList(clusteredVectors))
if(len(vecs) > 1):
last = vecs[-2]
mat = []
for atom1 in last:
row = []
for atom2 in clusteredVectors:
row.append(getDistanceBetweenAtoms(atom1, atom2))
mat.append(row)
mats.append(mat)
if(len(vecs)==len(atoms)):
next = vecs[0]
mat = []
for atom1 in clusteredVectors:
row = []
for atom2 in next:
row.append(getDistanceBetweenAtoms(atom1, atom2))
mat.append(row)
mats.append(mat)
return vecs, mats
def appendAtom(atom1, atom2):
atom1 = pe.toList(atom1)
atom2 = pe.toList(atom2)
last = (atom1[-1] + atom2[1])/2.0
beforeLast = (atom1[-2] + atom2[0])/2.0
return atom1[:-2] + [beforeLast, last] + atom2[2:]
