def computeKMoutliers(synthD):
num = 0
stats = []
for f in synthD:
#reads data and parses first file in folder
sd = readSynthetic(f)
print("Iteration:", num)
num += 1
for i in range(5):
#Running kMeansOut on the data
kcent, cid, dist = kmo.kmeansOutliers(sd.data,
compute_phi_star(sd), sd.z,
sd.k)
ans = kcent
kmo_cost, index_list = kmo.cost(sd.data, cid, kcent, sd.z)
average_cost = np.sum(kmo_cost)
#Computing cost
sd.costs.append(average_cost)
#example for adding extra stats, i.e. time. For headers, go to top
sd.extrastats = [0, num, num * num]
printSD(sd)
stats = addAnswer(stats, sd)
return stats
def computeKMOutliersLloyd(synthD):
num = 0
stats = []
for f in synthD:
#reads data and parses first file in folder
sd = readSynthetic(f)
sd.phistar = compute_phi_star(sd)
print("-------------------------\nIteration:", num,
"\n-------------------------")
num += 1
for rp in phistarprop:
for j in range(1):
sd.runphi = rp
sd.runk = sd.k + j
precs = []
recs = []
for i in range(10):
#Running kMeansOut on the data
centers, cid, dist, wins = kmo.kmeansOutliers(
sd.data, sd.phistar * sd.runphi, sd.z, sd.runk)
zind = []
for i in range(sd.k, sd.k + sd.z):
zind.append(i)
ans, cid, wins, prec, rec = lloyd.LloydOut(
sd.data, centers, sd.runk, sd.z, 1, 100, zind)
#kmo_cost, index_list = kmo.cost(sd.data, cid, ans, int(sd.z))
#average_cost= np.sum(kmo_cost)
cost2 = kmo.cost2(sd.data, ans, int(sd.z))
#print("Sharvaree_cost:", average_cost)
#assert(cost2 == average_cost)
#Computing cost
sd.costs.append(cost2)
precs.append(prec)
recs.append(rec)
sd.precs = precs
sd.recs = recs
#example for adding extra stats, i.e. time. For headers, go to top
sd.extrastats = [
sd.phistar,
mean(np.array(precs)),
max(precs),
mean(np.array(recs)),
max(recs)
]
print(sd.phistar)
printSD(sd)
stats = addAnswer(stats, sd)
sd.costs = []
return stats
def computeKMLS(synthD):
num = 0
stats = []
for f in synthD:
#reads data and parses first file in folder
sd = readSynthetic(f)
sd.phistar = compute_phi_star(sd)
print("Iteration:", num)
num += 1
for j in range(int(sd.k / 2)):
print("TrueCost:", sd.phistar)
sd.runphi = 1
sd.runk = sd.k + j
precs = []
recs = []
for i in range(1):
numpts = 2 * (sd.k + sd.z)
#numpts = int(sd.n/2)
sampleData = kmpp.kmeanspp(sd.data, numpts) #kmpp sampling
#sampleData = sd.data #No sampling, run on all
#sampleData = ls.randomInit(sd.data,numpts) #uniformly random init
#Running kcenterOut on the data
ans, empz = ls.lsOut(sampleData, sd.runk, sd.z, eps)
cost2 = kmo.cost2(sd.data, ans, int(sd.z))
#Computing cost
sd.costs.append(cost2)
prec, rec = km.kMPrecRecallVar2(sd, ans, empz)
precs.append(prec)
recs.append(rec)
sd.precs = precs
sd.recs = recs
#example for adding extra stats, i.e. time. For headers, go to top
sd.extrastats = [
sd.phistar,
mean(np.array(precs)),
max(precs),
mean(np.array(recs)),
max(recs)
]
printSD(sd)
stats = addAnswer(stats, sd)
sd.costs = []
return stats
def computeKMPP(synthD):
num = 0
stats = []
for f in synthD:
#reads data and parses first file in folder
sd = readSynthetic(f)
sd.phistar = compute_phi_star(sd)
print("Iteration:", num)
num += 1
for j in range(int(sd.k / 2)):
print("TrueCost:", sd.phistar)
sd.runphi = 1
sd.runk = sd.k + j
precs = []
recs = []
for i in range(10):
ans = kmpp.kmeanspp(sd.data, sd.runk)
cost2 = kmo.cost2(sd.data, ans, int(sd.z))
#Computing cost
sd.costs.append(cost2)
prec, rec = km.kMPrecRecall(sd, ans)
precs.append(prec)
recs.append(rec)
sd.precs = precs
sd.recs = recs
#example for adding extra stats, i.e. time. For headers, go to top
sd.extrastats = [
sd.phistar,
mean(np.array(precs)),
max(precs),
mean(np.array(recs)),
max(recs)
]
printSD(sd)
stats = addAnswer(stats, sd)
sd.costs = []
return stats
def computeKMLSCoreset(synthD):
num = 0
stats = []
for f in synthD:
#reads data and parses first file in folder
sd = readSynthetic(f)
sd.phistar = compute_phi_star(sd)
print("Iteration:", num)
num += 1
for j in range(int(sd.k / 2)):
print("TrueCost:", sd.phistar)
print(distance.cdist(sd.data[0:sd.k], sd.data[0:sd.k]))
sd.runphi = 1
sd.runk = sd.k + j
precs = []
recs = []
for i in range(1):
#Running kcenterOut on the data
numpts = 2 * (sd.k + sd.z)
print("numpts", numpts)
#numpts = int(sd.n/10)
#numpts = sd.k + sd.z
ans, empz = ls.lsOutCor(sd.data,
sd.runk,
sd.z,
eps,
numpts,
debug=True)
cost2 = kmo.cost2(sd.data, ans, int(sd.z))
print("------------\nBefore Lloyds:", cost2, empz)
print(find_inds(sd.data, ans))
prec, rec = km.kMPrecRecallVar2(sd, ans, int(sd.z))
print("Prec, rec", prec, rec)
cr = km.cr1(sd, ans)
zind = []
for i in range(sd.k, sd.k + sd.z):
zind.append(i)
ans, cid, wins, prec, rec, garbage = lloyd.LloydOut(
sd.data, ans, sd.runk, sd.z, 1, 100, zind)
cost2 = kmo.cost2(sd.data, ans, int(sd.z))
#Computing cost
sd.costs.append(cost2)
print("After Lloyds:", cost2)
print("Prec, rec", prec, rec)
cr = km.cr1(sd, ans)
input()
precs.append(prec)
recs.append(rec)
sd.precs = precs
sd.recs = recs
#example for adding extra stats, i.e. time. For headers, go to top
sd.extrastats = [
sd.phistar,
mean(np.array(precs)),
max(precs),
mean(np.array(recs)),
max(recs)
]
printSD(sd)
stats = addAnswer(stats, sd)
sd.costs = []
return stats
def compute_phi_star(sd):
return kmo.cost2(sd.data[sd.k + sd.z:], sd.data[:sd.k], int(sd.z))