def train(self, pos, neg, k=7, nofIterations=20, encoding="utf8"):
with open(pos, encoding=encoding) as fpos:
tweets_pos = fpos.readlines()
with open(neg, encoding=encoding) as fneg:
tweets_neg = fneg.readlines()
print("representing training data...")
tweets_pos = np.array(
[self.representation(tweet) for tweet in tweets_pos])
tweets_neg = np.array(
[self.representation(tweet) for tweet in tweets_neg])
print("fitting...")
self.__cent_pos = kmeans(tweets_pos, k=k, iterationCount=nofIterations)
self.__cent_neg = kmeans(tweets_neg, k=k, iterationCount=nofIterations)
ntotal = len(tweets_pos) + len(tweets_neg)
ncorrect = 0
self._clf = (self.__cent_pos, self.__cent_neg)
self._store_clf()
for t in tweets_pos:
if self._predict(t) == 1:
ncorrect += 1
for t in tweets_neg:
if self._predict(t) == -1:
ncorrect += 1
accuracy = ncorrect / ntotal
print("classifier trained")
print(f"accuracy on training set:{accuracy}")
def train_function(x, y, model, window_state):
if not model:
class Model:
centroids = None
k = self.k
sum_iterations = 0
sum_error = 0
i = 0
model = Model()
if model.centroids is not None and self.incremental:
[centroids, index, i] = kmeans(x, model.k, model.centroids,
draw=self.draw,
output=self.output)
else:
[centroids, index, i] = kmeans(x, model.k, draw=self.draw,
output=self.output)
model.centroids = centroids
self.centroids = centroids
error = evaluate_error(x, centroids, index)
if self.output:
print "Error: ", error
model.sum_iterations += i
model.sum_error += error
model.i += 1
return model
def kmeansMain():
kmax = 10
clustering = []
clustering.append(np.load('hw2_data/kmeans/clustering1.npy'))
clustering.append(np.load('hw2_data/kmeans/clustering2.npy'))
clustering.append(np.load('hw2_data/kmeans/clustering3.npy'))
clustering.append(np.load('hw2_data/kmeans/clustering4.npy'))
iterator = 1
bests = [2, 3, 4, 5]
for cluster in clustering:
transposed = np.transpose(cluster)
mins = []
maxs = []
for dim in transposed:
mins.append(min(dim))
maxs.append(max(dim))
bestObjs = []
for k in range(1, kmax + 1):
objs = []
for init in range(k):
centers = []
for i in range(k):
data = []
for j in range(len(mins)):
data.append(np.random.uniform(mins[j], maxs[j]))
centers.append(data)
_, obj = kmeans(cluster, np.array(centers))
objs.append(obj)
bestObjs.append(min(objs))
plt.plot(range(1, kmax + 1), bestObjs)
plt.ylabel("Value of Objective Function")
plt.xlabel("k")
plt.title("Cluster " + str(iterator))
plt.show()
k = bests[iterator - 1]
centers2 = []
for i in range(k):
data2 = []
for j in range(len(mins)):
data2.append(np.random.uniform(mins[j], maxs[j]))
centers2.append(data2)
resCenters, _ = kmeans(cluster, np.array(centers2))
assignments = assign_clusters(cluster, resCenters)
assigned = []
assigned2 = []
for i in range(k):
assigned.append([])
for i in range(len(cluster)):
assigned[assignments[i]].append(cluster[i])
for arr in assigned:
assigned2.append(np.array(arr))
plt.clf()
colors = ['or', 'ok', 'ob', 'og', 'oy']
for i in range(k):
plt.plot(assigned2[i][:, 0], assigned2[i][:, 1], colors[i])
plt.title("K-means Clusters")
plt.show()
iterator += 1
def clusterConvergence2Modes(xk, activeMeans):
# evaluate the fit of two means to the data
testMeans = 2
(idxk, mui) = kmeans.kmeans(xk.transpose(), testMeans)
# dimension of the data
p = xk.shape[0]
# numnber of particles
N = xk.shape[1]
# evaluate the mean and covariance of each cluster
meansk = np.zeros((p, testMeans))
Pkkk = np.zeros((p, p, testMeans))
# evaluate the mean and covariance of each cluster
for k in range(testMeans):
# compute the mean of all members where meansIdx == k
idx = np.nonzero(idxk == k)
idx = idx[0]
meansk[:, k] = mui[k, :].transpose()
# compute the covariance
Pkk = np.zeros((p, p))
coef = 1.0 / (float(N) - 1.0)
for j in idx:
Pkk = Pkk + coef * np.outer(xk[:, j] - meansk[:, k], xk[:, j] - meansk[:, k])
Pkkk[:, :, k] = Pkk.copy()
# evaluate the likelihood for each point, under the bimodal assumption
L2 = np.zeros(N)
for k in range(N):
# assume the likelihood is proportional to the PDF
pxk = gaussianNormalPdf(xk[:, j], meansk[:, idxk[k]], Pkkk[:, :, idxk[k]])
L2[k] = pxk
# Akaike information criterion for bimodal case
AIC2 = 2.0 * 4 - 2.0 * math.log(np.max(L2))
# evaluate the likelihood under the monomodal assumption
P11 = np.zeros((p, p))
mux = np.mean(xk, axis=1)
for k in range(N):
P11 = P11 + (1.0) / (float(N) - 1.0) * np.outer(xk[:, k] - mux, xk[:, k] - mux)
L = np.zeros(N)
for k in range(N):
L[k] = gaussianNormalPdf(xk[:, j], mux, P11)
# information criterion for unimodal case
AIC = 2.0 * 2 - 2.0 * math.log(np.max(L))
# smaller AIC value is better
print("AIC1 = %f,AIC2 = %f, L1 = %f, L2 = %f" % (AIC, AIC2, np.max(L), np.max(L2)))
# if AIC < AIC2:
if np.max(L) > np.max(L2):
(idxk, mui) = kmeans.kmeans(xk.transpose(), 1)
return (idxk, mui)
def get(self):
url = self.request.get('url')
k = int(self.request.get('k', default_value=3))
t = int(self.request.get('t', default_value=144))
result = urlopen(url)
# Get image
if (result.getcode() == 200):
data = result.read()
img = images.Image(data)
raw_width = img.width
raw_height = img.height
width = raw_width
height = raw_height
# Resize to max
if raw_width > MAX and raw_height > MAX:
if raw_width > raw_height:
height = MAX
width = width * MAX / height
else:
width = MAX
height = height * MAX / width
img = images.resize(data, width, height)
reader = png.Reader(bytes=img)
# Get pixels
(width, height, pixels, meta) = reader.read_flat()
n = PIXEL_SIZE
(means, counts) = kmeans.kmeans(k, pixels, n, 255, t)
colors = [hex_string(means[i * n : i * n + n]) for i in range(len(counts))]
output = {}
output['colors'] = [{'color': colors[i], 'count': counts[i]} for i in range(len(counts))]
output['sum'] = sum(counts)
self.response.headers['Content-Type'] = 'application/json'
self.response.write(json.dumps(output))
def _init(X, mean, diagonol=False): km = kmeans(K) km.means = mean km.assign_clusters(X) m, sigma, pi = _extract_from_kmeans(km) sigma = _fix_sigma(sigma, def_sigma_tol) return sigma, pi
def vect_quant(input_file_folder, input_file_name, X, means_file, K):
# print('Launching Thread vect_quant for ', input_file_folder, input_file_name, K)
f = open(means_file + 'K' + str(K) + '.txt', 'r')
lines = f.readlines()
f.close()
f = None
w = []
for point in lines:
w.append(np.fromstring(point, dtype=float, sep=' '))
means = np.array(w)
w = None
point = None
lines = None
gc.collect()
friday = kmeans(K)
friday.means = means
friday.initialize_clusters(X)
friday.assign_clusters(X)
means = None
gc.collect()
clusters = friday.clusters
firday = None
gc.collect()
output_folder = input_file_folder + '/vq' + str(K)
if not os.path.exists(output_folder):
os.makedirs(output_folder)
output_file = open(output_folder + '/' + input_file_name[:-4] + '.txt',
'w')
for c in clusters:
output_file.write(str(c) + ' ')
def main():
# Input file that contains the input for
# commuters, cabs and destination location
input_file = "sample_inputs/%s" % SAMPLE_INPUT_FILE
if len(sys.argv) == 2:
input_file = sys.argv[1]
# Parse input file
commuters, cabs, destination = parse_input_file(input_file)
# Create clusters of commuters
clusters = kmeans(commuters, cabs)
# Centroid of each cluster represents virtual centre of
# each commuter group
groups = [c.centroid for c in clusters]
# Find distance between each commuter group and cab
all_distances = get_group_cab_distances(groups, cabs)
# Find optimal total distance and route travelled by all the cabs
optimal_distance, optimal_route = optimal_total_distance(
all_distances, groups, cabs)
# Add total distance between group and destination to optimal distance
optimal_distance = add_destination_distance(optimal_distance,
groups, destination)
# Print answer
print_answer(optimal_distance, optimal_route, clusters, cabs)
def find_k_star(data_points, threshold):
""" Starting with 1 cluster, finds out the optimum value of k*.
In each iteration, the value of number of clusters is doubled.
"""
num_clusters = 1
prev_cohesion = None
while True:
clusters, cohesion = kmeans(data_points, num_clusters)
if prev_cohesion:
change_rate = float(abs(cohesion - prev_cohesion)) / (
prev_cohesion * num_clusters / 2)
if change_rate < threshold:
break
prev_cohesion = cohesion
num_clusters *= 2
if num_clusters > len(data_points):
print len(data_points)
return
# perform binary search to find kstar. As num of clusters have already been doubled in previous for loop
# so start corresponds to num_clusters/4 and end to num_clusters/2
kstar = binary_search(data_points, num_clusters / 4, num_clusters / 2,
threshold)
print kstar
def KmeansCluster(FileName = "",Num = 3):
dataSet = []
fileIn = open(FileName)
for line in fileIn.readlines():
lineArr = line.strip().split(',')
dataSet.append([float(i) for i in lineArr])
dataSet = mat(dataSet)
k = Num
centroids, clusterAssment = km.kmeans(dataSet, k)
return centroids,clusterAssment
##Demo
# FileName = "/Users/nevin47/Desktop/项目/学术/TopTen_Python/DataMiningTopTen/KMeans/Table/FINAL.csv"
# Center,Out = KmeansCluster(FileName,3)
# # print Out
#
# dataSet = []
# fileIn = open(FileName)
# for line in fileIn.readlines():
# lineArr = line.strip().split(',')
# dataSet.append([float(i) for i in lineArr])
# dataSet = mat(dataSet)
# # C = transpose(mat(Center[1]))
# # D = dataSet[1]
# # print D*C
# Result = km.CalVectorDistance(dataSet,Out)
# # print Result[:,0,:]
# SUM = 0
# for i in range(dataSet.shape[0]):
# SUM += km.CalVectorCoefficient(Result,i,3)
# print SUM/dataSet.shape[0]
# # print km.CalVectorCoefficient(Result,0,3)
def part1_q4(dataset, show_graph=False):
print(">>> PART 1: QUESTION 4")
clustering = kmeans(dataset, 2, initCentroids=[('i1', 3, 2), ('i2', 4, 8)], distance_type='Manhattan')
printTable(clustering["centroids"])
if show_graph:
showClusters2D(clustering)
print('')
def segment(self):
algorithm = self.algorithmComboBox.currentText()
imgPath = str(self.userpath.text())
index = self.featureComboBox.currentIndex()
features = ["INTENSITY","INTENSITY+LOC","RGB","YUV","LM",
"ILM","PCA"]
if algorithm == "K-means":
k = int(self.kText.text())
iterations = int(self.iterationsText.text())
epsilon = float(self.epsilonText.text())
print imgPath,index,k,iterations,epsilon
org = cv.LoadImageM(imgPath)
im = kmeans.kmeans(imgPath,features[index],k,iterations,epsilon)
cv.ShowImage("original",org)
cv.ShowImage("segmented",im)
elif algorithm == "Mean Shift":
if index == 4:
QtGui.QMessageBox.information(self, 'Error',
'LM is not supported in mean shift')
return
print imgPath,features[index]
org = cv.LoadImageM(imgPath)
im = meanshift.meanshift(imgPath,features[index])
cv.ShowImage("original",org)
cv.ShowImage("segmented",im)
def part1_q2(dataset, show_graph=False):
print(">>> PART 1: QUESTION 2")
clustering = kmeans(dataset, 2, initCentroids=[('i1', 4, 6), ('i2', 5, 4)], distance_type='Euclidean')
printTable(clustering["centroids"])
if show_graph:
showClusters2D(clustering)
print('')
def kmeans_test(request):
data = None
form = kmeansNumSamplesForm()
k=None
sample_size=None
grouped_data = None
clusters = None
error_list = None
if request.method=='POST':
form = kmeansNumSamplesForm(request.POST)
if form.is_valid():
sample_size = int(form.cleaned_data['num_samples'])
k = int(form.cleaned_data['k'])
# Generate random data
data = numpy.random.random((sample_size, 2))
# Calculate kmeans
if form.cleaned_data['method']=='Basic':
grouped_data, clusters, error_list = kmeans.kmeans(data,num_clusters=k, min_error=0.01, max_iter=100)
else:
grouped_data, clusters, error_list = kmeans.bisecting_kmeans(data,k=k, min_error=0.01, max_iter=50)
return render_to_response('visualization/kmeans.html', {
'data': grouped_data,
'clusters': clusters,
'error_list': error_list,
'form':form,
'k': k,
'sample_size': sample_size,
}, context_instance=RequestContext(request))
def main():
import matplotlib.pyplot as plt
from sklearn.datasets.samples_generator import make_blobs
n_centers = 3
X, y = make_blobs(n_samples=1000, centers=n_centers, n_features=2,
cluster_std=0.7, random_state=0)
# Run this K-Means
import kmeans
t0 = time.time()
y_pred, centers, obj_val_seq = kmeans.kmeans(X, n_centers)
t1 = time.time()
print("Final obj val: {}".format(obj_val_seq[-1]))
print("Time taken (this implementation): {}".format(t1 - t0))
# Run scikit-learn's K-Means
from sklearn.cluster import k_means
t0 = time.time()
centers, y_pred, obj_val = k_means(X, n_centers, random_state=0)
t1 = time.time()
print("Final obj val: {}".format(obj_val))
print("Time taken (Scikit, 1 job): {}".format(t1 - t0))
# Plot change in objective value over iteration
fig = plt.figure()
ax = fig.add_subplot(111)
ax.plot(obj_val_seq, 'b-', marker='*')
fig.suptitle("Change in K-means objective value across iterations")
ax.set_xlabel("Iteration")
ax.set_ylabel("Objective value")
fig.show()
# Plot data
from itertools import cycle
colors = cycle('bgrcmykbgrcmykbgrcmykbgrcmyk')
fig = plt.figure(figsize=plt.figaspect(0.5)) # Make twice as wide to accomodate both plots
ax = fig.add_subplot(121)
ax.set_title("Data with true labels and final centers")
for k, color in zip(range(n_centers), colors):
ax.plot(X[y==k, 0], X[y==k, 1], color + '.')
initial_centers = kmeans.init_centers(X, n_centers, 2) # This is valid because we always use the same random seed.
# Plot initial centers
for x in initial_centers:
ax.plot(x[0], x[1], "mo", markeredgecolor="k", markersize=8)
# Plot final centers
for x in centers:
ax.plot(x[0], x[1], "co", markeredgecolor="k", markersize=8)
# Plot assignments
colors = cycle('bgrcmykbgrcmykbgrcmykbgrcmyk')
ax = fig.add_subplot(122)
ax.set_title("Data with final assignments")
for k, color in zip(range(n_centers), colors):
ax.plot(X[y_pred==k, 0], X[y_pred==k, 1], color + '.')
fig.tight_layout()
fig.gca()
fig.show()
def main(filename, k, function):
file = open(filename)
pts = []
for line in file.readlines():
line = line.split()
pts.append(( int(line[0]), int(line[1])))
file.close()
group, s = kmeans.kmeans(pts, k, function)
print group
print s
f = open(filename +'cluster', 'w')
for each in range(len(group)):
f.write( str(group[each])+'\n\n')
cc = numpy.random.rand(10000)
for i in group[each]:
matplotlib.pyplot.scatter(i[0],i[1],s=2, color=cc)
pylab.plot(s[each][0], s[each][1], '+')
f.close()
ll = filename + 'cluster'
ll = 'gedit '+ ll
ll = ll.split()
process = Popen(ll)
matplotlib.pyplot.show()
def main():
finput = open('../resources/BB.txt','r')
mat = np.loadtxt(finput,delimiter=' ',)
finput.close()
X = prepareX()
k = 10
min_samples = 4
opt_cutoff = 0.5
threshold = 500
reduce_threshold = 600
est = kmeans(X,k,opt_cutoff)
est2 = kmeans2(X,opt_cutoff,threshold,reduce_threshold)
(cluster_label0,id2point) = buildLabels(est,X)
(cluster_label01,id2point) = buildLabels(est2,X)
cluster_label1 = SpectralClustering(k).fit_predict(mat)
cluster_label1 = buildLabels2(cluster_label1,X)
cluster_label2 = DBSCAN(min_samples=min_samples).fit_predict(mat)
cluster_label2 = buildLabels2(cluster_label2,X)
np.savetxt('../resources/KMeans.txt', cluster_label0, fmt='%s', newline='\n', header='', footer='', comments='# ')
np.savetxt('../resources/KMeans2.txt', cluster_label01, fmt='%s', newline='\n', header='', footer='', comments='# ')
np.savetxt('../resources/SpectralClustering.txt', cluster_label1, fmt='%s', newline='\n', header='', footer='', comments='# ')
np.savetxt('../resources/DBSCAN.txt', cluster_label2, fmt='%s', newline='\n', header='', footer='', comments='# ')
np.savetxt('../resources/ID2Point.txt', id2point, fmt=["%s",]*3, newline='\n')
def clustering():
# Create pairs for clustering (age/eng/prog/uni)
pairs = []
for i, val in enumerate(uni):
if(val > 0 and prog[i] > 0):
pairs.append((val, prog[i]))
# Run k-means
print(pairs)
seeds = [pairs[0], pairs[1], pairs[2]]
clusters = kmeans.kmeans(pairs, seeds)
# Show diagram with coloured clusters
for i in clusters:
cluster = clusters[i]
xaxis = []
yaxis = []
for pair in cluster:
xaxis.append(pair[0])
yaxis.append(pair[1])
print(yaxis)
plt.plot(xaxis, yaxis, "o")
plt.axis([0,12,0,12])
plt.xlabel('University years')
plt.ylabel('Programming skill')
plt.show()
def __init__(self,
inputs,
targets,
nRBF,
sigma=0,
usekmeans=0,
normalise=0):
self.nin = inputs.shape[1]
self.nout = targets.shape[1]
self.ndata = inputs.shape[0]
self.nRBF = nRBF
self.usekmeans = usekmeans
self.normalise = normalise
if usekmeans:
self.kmeansnet = kmeans.kmeans(self.nRBF, inputs)
self.hidden = np.zeros((self.ndata, self.nRBF + 1))
if sigma == 0:
d = (inputs.max(axis=0) - inputs.min(axis=0)).max()
self.sigma = d / np.sqrt(2 * nRBF)
else:
self.sigma = sigma
self.perceptron = pcn.pcn(self.hidden[:, :-1], targets)
self.weights1 = np.zeros((self.nin, self.nRBF))
def sc(data_points: ndarray, k: int) -> (ndarray, list):
"""
sc(data_points, k)
Run the spectral clustering
Parameters
----------
data_points : ndarray
Data points to be clustered.
k : int
number of clusters
Returns
-------
(ndarray, list)
cc: The centroids of clusters
aff: The affectation of each node to it's cluster
if aff[i] = j then the node i is in the cluster j
"""
logging.info("Starting Spectral Clustering")
# Calculate the eigenvectors
logging.info("Calculating the eigenvectors")
_, u = la.eigh(data_points, eigvals=(0, k))
# Run the k-means
logging.info("Running k-means")
# cc, aff = cl.vq.kmeans2(u, k)
cc, aff = kmeans(u, k)
return cc, aff
def questao21():
dset = load_dataset('dataset1.csv')
xo = dset.T[1].astype(float) # segunda coluna
x = dset.T[1].astype(float) # segunda coluna
yo = dset.T[2].astype(float) # terceira coluna
y = dset.T[2].astype(float) # terceira coluna
# a normalização com z-score ajudou na visualização e é necessária para agrupamento
#x = [z_score(x, xi) for xi in x]
#y = [z_score(y, yi) for yi in y]
#centros_iniciais = [(z_score(xo, 1), z_score(yo, 2)), (z_score(xo, 4), z_score(yo, 2))]
centros_iniciais = [(1, 2), (4, 2)]
pontos = zip(x, y)
clusters, iteracoes = kmeans(pontos, 2, centros_iniciais=centros_iniciais)
cluster1 = clusters[0].pontos
cluster2 = clusters[1].pontos
plt.plot([xi[0] for xi in cluster1], [yi[1] for yi in cluster1], 'ro')
plt.plot([clusters[0].centroide[0]], [clusters[0].centroide[1]], 'r*')
plt.plot([xi[0] for xi in cluster2], [yi[1] for yi in cluster2], 'go')
plt.plot([clusters[1].centroide[0]], [clusters[1].centroide[1]], 'g*')
plt.savefig('grupo1.png')
print "Novos centróides:", clusters[0].centroide, " e ", clusters[
1].centroide
def main():
try:
inputFile = datFile = sys.argv[1]
k = sys.argv[2]
except Exception as e:
print('Oh No! => %s' % e)
print('Usage:\npython3 ./main.py <data.mat> <k>')
sys.exit(2)
mat = spio.loadmat(inputFile, squeeze_me=True)
fname = os.path.splitext(inputFile)
rawdata = mat[fname[0]]
print(fname[0])
dataIn = rawdata[:, (0, 1)]
mu, clusters = km.kmeans(int(k), 10, 0.00000001, dataIn)
colors = ["r", "b", "g", "k", "c", "y", "m"]
for i in range(0, max(clusters) + 1):
indices = myFind(clusters, lambda x: x == i)
x = dataIn[indices, 0]
y = dataIn[indices, 1]
plt.scatter(x, y, c=colors[i])
plt.scatter(mu[i, 0], mu[i, 1], c="yellow", marker="*", s=300)
plt.title("K-means results")
plt.show()
def test_kmeans():
name = "input-test-kmeans.txt"
out = kmeans.kmeans(name)
#print "gmis result = ", out
#note: all the indexest of the graph above are different from gmis, as here we start at 0,
#but gmis starts at 1, which means that we have a shift for all nodes
expectedresult1 = [
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1
]
expectedresult2 = [
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2
]
failed = False
for i in xrange(len(expectedresult1)):
if out[i] != expectedresult1[i]:
failed = True
if failed:
for i in xrange(len(expectedresult2)):
if out[i] != expectedresult2[i]:
failed = True
if failed:
sys.stderr.write(
"Possible problem in test_kmeans(). Might only be a different random output typical of k-means: "
+ str(out) + "\n")
def main():
# Read in the dataset
X = np.loadtxt('data/fisher_iris_data.csv', delimiter=',').T
Y = np.loadtxt('data/fisher_iris_labels.csv', dtype=str)
labels = list(set(Y)) # get unique labels
num_classes = len(labels)
Y = [labels.index(y) for y in Y] # convert labels to integers
## Plot dataset
fig = plt.figure()
plt.subplot(2,2,1)
plt.scatter(X[0, :], X[1, :])
plt.title('Data without labels')
plt.subplot(2,2,2)
plt.scatter(X[0, :], X[1, :], c=Y)
plt.title('Data with true labels')
ax1 = plt.subplot(2,2,1)
ax2 = plt.subplot(2,2,4)
scatterPlot = 0
num_evals = 100
kmeans_performance = np.zeros(num_evals)
kmeans_solutions = []
for i in range(num_evals):
print('\nK-Means Run ', i+1)
cluster_idx = kmeans(X, num_classes, ax1=ax1, ax2=ax2)
kmeans_performance[i] = evaluate(X, Y, cluster_idx)
print('Accuracy: ', kmeans_performance[i])
kmeans_solutions.append(cluster_idx)
plt.subplot(2, 2, 3)
plt.hist(kmeans_performance)
def sampleClustersGraph():
data, featureNames = load_2d_data()
k = 6
iniCenters = orderedCenters(data, k)
clusters, centers = kmeans(data, k,
initialCenters=iniCenters)
graphClusters(clusters, centers, data, featureNames)
def train(self, X):
"""
Compute K-Means clustering on each class label and store your result in self.cluster_centers_
:param X: inputs of training data, a 2D Numpy array
:return: None
"""
self.cluster_centers_ = kmeans(X, self.k, self.max_iter, self.tol)
def __init__(self,inputs,targets,nRBF,sigma=0,usekmeans=0,normalise=0):
self.nin = shape(inputs)[1]
self.nout = shape(targets)[1]
self.ndata = shape(inputs)[0]
self.nRBF = nRBF
self.usekmeans = usekmeans
self.normalise = normalise
#print "Initalizing RBFN with parameters: "
#print "Inputs : " + str(shape(inputs))
#print "targets : " + str(shape(targets))
#print "nRBF : " + str(nRBF)
# print "Sigma : " + str(sigma)
# print "K-Means : " + str(usekmeans)
# print "Normalise: " + str(normalise)
# print
if usekmeans:
self.kmeansnet = kmeans.kmeans(self.nRBF,inputs)
self.hidden = zeros((self.ndata,self.nRBF+1))
if sigma==0:
# Set width of Gaussians
d = (inputs.max(axis=0)-inputs.min(axis=0)).max()
self.sigma = d/sqrt(2*nRBF)
else:
self.sigma = sigma
self.perceptron = pcn.pcn(self.hidden[:,:-1],targets)
# Initialise network
self.weights1 = zeros((self.nin,self.nRBF))
def cluster_data(inputpath, outputpath, index_column, name_column,
columns_to_load, num_clusters, max_iterations):
names, input_data = read_and_normalise_csv(inputpath, index_column,
name_column, columns_to_load)
cluster_map = kmeans(input_data, num_clusters, max_iterations)
f = None
try:
f = open(outputpath, 'w')
except FileNotFoundError:
try:
f = open(outputpath, 'x')
except:
print("ERROR: could not load file")
return (0, [], 0)
#cluster_ary = sorted(list(zip(names, cluster_map)), key = lambda x: x[1], reverse=True)
cluster_ary = [[] for c in range(num_clusters)]
for i in range(len(names)):
cluster_ary[cluster_map[i]].append(names[i])
f.write(json.dumps(cluster_ary))
f.close()
return cluster_ary
def __init__(self, topic_dict, vocab_list, k = 4):
print("create cluster model")
self.k = k
self.vocab_list = vocab_list
self.tar_list = self.__dict2dataInfo(topic_dict, vocab_list)
print("K means ...")
[self.clusters, self.centroids] = kmeans.kmeans(self.tar_list, k)
def spectral(W, k):
'''
SPECTRUAL spectral clustering
Input:
W: Adjacency matrix, N-by-N matrix
k: number of clusters
Output:
idx: data point cluster labels, n-by-1 vector.
'''
# YOUR CODE HERE
n, n = np.shape(W)
idx = np.zeros((n ,1))
D = np.zeros((n, n))
for i in range(n):
D[i][i] = np.sum(W[i][:])
L = D - W
# begin answer
eng = np.linalg.eig(L)
enval = eng[0]
en = eng[1]
sort_index = np.argsort(enval)
topk = en[:,sort_index[:k]]
idx = kmeans(topk, k)
return idx
def KmeansCluster(FileName="", Num=3):
dataSet = []
fileIn = open(FileName)
for line in fileIn.readlines():
lineArr = line.strip().split(',')
dataSet.append([float(i) for i in lineArr])
dataSet = mat(dataSet)
k = Num
centroids, clusterAssment = km.kmeans(dataSet, k)
return centroids, clusterAssment
##Demo
# FileName = "/Users/nevin47/Desktop/项目/学术/TopTen_Python/DataMiningTopTen/KMeans/Table/FINAL.csv"
# Center,Out = KmeansCluster(FileName,3)
# # print Out
#
# dataSet = []
# fileIn = open(FileName)
# for line in fileIn.readlines():
# lineArr = line.strip().split(',')
# dataSet.append([float(i) for i in lineArr])
# dataSet = mat(dataSet)
# # C = transpose(mat(Center[1]))
# # D = dataSet[1]
# # print D*C
# Result = km.CalVectorDistance(dataSet,Out)
# # print Result[:,0,:]
# SUM = 0
# for i in range(dataSet.shape[0]):
# SUM += km.CalVectorCoefficient(Result,i,3)
# print SUM/dataSet.shape[0]
# # print km.CalVectorCoefficient(Result,0,3)
def getDepht(self, im3d, bbox):
x1 = int(bbox[0])
y1 = int(bbox[1])
x2 = int(bbox[2])
y2 = int(bbox[3])
average = 0.0
strIm = ""
num = 0
filteredImage = {}
for y in range(y1, y2 + 1):
strIm += "|"
for x in range(x1, x2 + 1):
d = im3d[y][x][0]
strIm += "{:>3}|".format(d)
#if d<=7: continue
filteredImage[(y, x)] = d
average += 1.0 * d
num += 1
strIm += "\n"
#print strIm
c1, c2 = kmeans(filteredImage, 2, 1000, 320, 240)
depth1 = average / num
depth2 = min([c1, c2])
depth1 = (depth1 + 18.7579) / 0.5181
depth2 = (depth2 + 18.7579) / 0.5181
print depth1
print depth2
if abs(depth1 - depth2) > 10:
depth = depth2
else:
depth = depth1
return depth
def questao21():
dset = load_dataset('dataset1.csv')
xo = dset.T[1].astype(float) # segunda coluna
x = dset.T[1].astype(float) # segunda coluna
yo = dset.T[2].astype(float) # terceira coluna
y = dset.T[2].astype(float) # terceira coluna
# a normalização com z-score ajudou na visualização e é necessária para agrupamento
#x = [z_score(x, xi) for xi in x]
#y = [z_score(y, yi) for yi in y]
#centros_iniciais = [(z_score(xo, 1), z_score(yo, 2)), (z_score(xo, 4), z_score(yo, 2))]
centros_iniciais = [(1,2), (4,2)]
pontos = zip(x, y)
clusters, iteracoes = kmeans(pontos, 2, centros_iniciais=centros_iniciais)
cluster1 = clusters[0].pontos
cluster2 = clusters[1].pontos
plt.plot([xi[0] for xi in cluster1], [yi[1] for yi in cluster1], 'ro')
plt.plot([clusters[0].centroide[0]], [clusters[0].centroide[1]], 'r*')
plt.plot([xi[0] for xi in cluster2], [yi[1] for yi in cluster2], 'go')
plt.plot([clusters[1].centroide[0]], [clusters[1].centroide[1]], 'g*')
plt.savefig('grupo1.png')
print "Novos centróides:", clusters[0].centroide, " e ", clusters[1].centroide
def __init__(self,
inputs,
targets,
nRBF,
sigma=0,
usekmeans=0,
normalise=0):
self.nin = shape(inputs)[1]
self.nout = shape(targets)[1]
self.ndata = shape(inputs)[0]
self.nRBF = nRBF
self.usekmeans = usekmeans
self.normalise = normalise
if usekmeans:
self.kmeansnet = kmeans.kmeans(self.nRBF, inputs)
self.hidden = zeros((self.ndata, self.nRBF + 1))
if sigma == 0:
# Set width of Gaussians
d = (inputs.max(axis=0) - inputs.min(axis=0)).max()
self.sigma = d / sqrt(2 * nRBF)
else:
self.sigma = sigma
self.perceptron = pcn.pcn(self.hidden[:, :-1], targets)
# Initialise network
self.weights1 = zeros((self.nin, self.nRBF))
def main(algorithm, data, cl_labels, min_k, max_k, max_iterations, epsilon):
results, silhouette, chs, ssws, ssbs, ars, hom, comp = [], [], [], [], [], [], [], []
membership, centroids, labels = [], [], []
for c in range(min_k, max_k + 1):
if algorithm == 'kmeans':
labels, centroids = kmeans.kmeans(data, c)
elif algorithm == 'bisecting_kmeans':
labels, centroids = bisecting_kmeans.bisecting_kmeans(data, c)
elif algorithm == 'fuzzy_cmeans':
membership, centroids = fuzzyCmeans.execute(data, max_iterations, c, epsilon)
labels = fuzzyCmeans.get_labels(len(data), membership)
silhouette.append((c, metrics.silhouette_score(data, labels, metric='euclidean')))
chs.append((c, metrics.calinski_harabaz_score(data, labels)))
ssws.append((c, utils.get_ssw(data, centroids, labels)))
ssbs.append((c, utils.get_ssb(centroids)))
ars.append((c, metrics.adjusted_rand_score(cl_labels, labels)))
hom.append((c, metrics.homogeneity_score(cl_labels, labels)))
comp.append((c, metrics.completeness_score(cl_labels, labels)))
results.append(("Silhouette", "", zip(*silhouette)[0], "", zip(*silhouette)[1], 333, "blue"))
results.append(("Calinski-Harabaz Index", "", zip(*chs)[0], "", zip(*chs)[1], 334, "blue"))
results.append(("Intra cluster Variance", "", zip(*ssws)[0], "", zip(*ssws)[1], 331, "blue"))
results.append(("Inter cluster Variance", "", zip(*ssbs)[0], "", zip(*ssbs)[1], 332, "blue"))
results.append(("Adjusted Rand Index", "", zip(*ars)[0], "", zip(*ars)[1], 335, "orange"))
results.append(("Homogeneity", "", zip(*hom)[0], "", zip(*hom)[1], 336, "orange"))
results.append(("Completeness", "", zip(*comp)[0], "", zip(*comp)[1], 337, "orange"))
print(labels)
utils.plot_results(results, algorithm)
def __init__(self, topic_dict, vocab, k=4):
print("create cluster model")
self.k = k
self.vocab_info = [vocab, len(list(vocab.keys()))]
self.tar_list = self.__dict2dataInfo(topic_dict)
print("K means ...")
[self.clusters, self.centroids] = kmeans.kmeans(self.tar_list, k)
def initialize_population(self):
points = self.generate_random_array()
clusters, centroids = kmeans.kmeans(points, self.k)
initial_population = []
for i in range(1, self.k + 1):
tmp_points = points[clusters == i]
costs = [self.cf.fitness(point) for point in tmp_points]
max_idx = min(len(costs), self.pop_size // self.k)
best_indexes = sorted(range(len(costs)),
key=lambda i: costs[i])[:max_idx]
[
initial_population.append(tmp_points[idx])
for idx in best_indexes
]
if len(initial_population) < self.pop_size:
initial_population = np.append(
initial_population,
self.generate_random_array(
self.pop_size - len(initial_population), self.ind_size),
axis=0)
return np.array(initial_population)
def test_kmeans_9(self):
dataset = self.__load_dataset()
out = kmeans(dataset, 9)
percentage = avg_iou(dataset, out)
np.testing.assert_almost_equal(percentage, 0.672, decimal=2)
def clusterEigenvectors(k, laplacian, maxIterations):
#print
#print "---Eigenvector Clustering---"
# Call kmeans to cluster the resulting eigenvectors
clusters = kmeans(k, laplacian, maxIterations)
return clusters
def spectral(W, k):
'''
SPECTRUAL spectral clustering
Input:
W: Adjacency matrix, N-by-N matrix
k: number of clusters
Output:
idx: data point cluster labels, n-by-1 vector.
'''
# YOUR CODE HERE
# begin answer
N = W.shape[0]
D = (np.array(np.sum(W, axis=1)).T)[0]
#D = np.array(np.sum(W, axis=1))
L = np.diag(D) - W
# DLD
D_ = np.diag(1.0 / np.sqrt(D))
L = np.dot(np.dot(D_, L), D_)
value, vector = np.linalg.eig(L)
print(vector.shape)
value = zip(value, range(N))
value = sorted(value, key=lambda x: x[0])
a, b = value[1]
#H = vector[:, 1]
H = (np.array(vector[:, b]).T)[0]
t1 = np.mean(H)
t2 = np.std(H)
H = (H - t1) / t2
H = np.array([H]).T
res = kmeans(H, 2)
return res
def _get_anchors(self,
bboxes_in,
input_shape=(224, 224),
clusters=5,
strip_size=32):
'''
@input_shape tuple (h, w)
@bboxes_in format: [ [[xmin,ymin, xmax, ymax, label],], ]
value range: x [0, w], y [0, h]
@return anchors, format: 10 value tuple
'''
w = input_shape[1]
h = input_shape[0]
# TODO: add position to iou, not only box size
bboxes = []
for items in bboxes_in:
for bbox in items:
bboxes.append(
((bbox[2] - bbox[0]) / w, (bbox[3] - bbox[1]) / h))
bboxes = np.array(bboxes)
self.log.i(f"bboxes num: {len(bboxes)}, first bbox: {bboxes[0]}")
out = kmeans.kmeans(bboxes, k=clusters)
iou = kmeans.avg_iou(bboxes, out) * 100
self.log.i("bbox accuracy(IOU): {:.2f}%".format(iou))
self.log.i("bound boxes: {}".format(",".join(
"({:f},{:.2f})".format(item[0] * w, item[1] * h) for item in out)))
for i, wh in enumerate(out):
out[i][0] = wh[0] * w / strip_size
out[i][1] = wh[1] * h / strip_size
anchors = list(out.flatten())
self.log.i(f"anchors: {anchors}")
ratios = np.around(out[:, 0] / out[:, 1], decimals=2).tolist()
self.log.i("w/h ratios: {}".format(sorted(ratios)))
return anchors
def train_model(k=2):
#Train k-Means on the training data
#k=2
model = kmeans.kmeans(n_clusters=k)
model.fit(Xtrain)
#Predict back the training ratings and compute the RMSE
XtrainHat = model.predict(Xtrain,Xtrain)
tr= model.rmse(Xtrain,XtrainHat)
#Predict the validation ratings and compute the RMSE
XvalHat = model.predict(Xtrain,Xval)
val= model.rmse(Xval,XvalHat)
#Predict the test ratings and compute the RMSE
XtestHat = model.predict(Xtrain,Xtest)
te= model.rmse(Xtest,XtestHat)
#Get the cluster assignments for the training data
z = model.cluster(Xtrain)
print(z) , len(z)
#Get the clusters
centers = model.get_centers()
print(centers)
print("K=%d Errors: %.7f %.7f "%(k,tr,val))
def gmm_init(k, samples):
"""
init a gauss mixture model for all samples
using kmeans algorithm
weights don't sum up to 1
"""
centers = km.kmeans(k, samples)
clusters = km.cluster(samples, centers)
#params is a list of (mean, sigma, weight)
# shapec = np.shape(centers[0])
shapes = np.shape(np.outer(samples[0], samples[0]))
#params = [[np.zeros_like(centers[0]), np.zeros(shapes), 0]]*k
params = [None] * k
for i in range(k):
cluster, center = clusters[i], centers[i]
num_samples = len(cluster)
deviation = np.zeros(shapes)
for sample in cluster:
diff = sample - center
deviation += np.outer(diff, diff)
deviation /= len(cluster)
params[i] = [center, deviation, num_samples]
return params
def ikmeans(data_points):
"""
Parameters
----------
data_points: np.ndarray
The data points to be clustered
Returns
-------
(np.ndarray, np.ndarray)
2-D array of centroids
1-D array of affectation list of data nodes
"""
n = len(data_points)
logging.info("Starting the modified Spectral Clustering")
# Calculate the eigenvectors
logging.info("Calculating the eigenvectors")
_, u = la.eigh(data_points)
k = 3
cc = aff = None
# run the iterations
while True:
_u = u[:, :k]
centroids = km.get_random_initial_centroids(_u, k)
cg = np.average(_u, axis=0)
cg.resize((1, len(cg)))
centroids = np.concatenate((centroids, cg), axis=0)
cc, aff = km.kmeans(_u, k, centroids)
if len(np.unique(aff)) == k or k == n:
break
else:
k += 1
return cc, aff
def MNIST_eval_euclidean(metric_func, numbers=[1,2,3],
nrange=range(10,100,10), num_avg=10):
"""Return metric evaluation on MNIST dataset using Euclidean distance
on all the algorithms.
Input
-----
metric_func - metric being evaluated
numbers - digits chosen in MNIST data set
nrange - range of N's to be tested, number of data points
num_avg - number of times we cluster the same points and take
the average, min, and max
Output
------
kmedoids_metric - metric computed with K-medoids
kmeans_metric - metric computed with K-means
kmeans_sklearn_metric - metric with kmeans from sklearn
"""
digits = datasets.load_digits()
images = digits.images
kmedoids_metric = []
kmeans_metric = []
kmeans_sklearn_metric = []
for n in nrange:
# generate true labels
labels = np.concatenate([[m]*n for m in numbers])
data = np.concatenate([
images[np.where(digits.target==i)][np.random.choice(range(173), n)]
for i in numbers
])
data2 = data.reshape(len(data), 64)
m1 = []; m2 = []; m3 = [];
for i in range(num_avg):
# our algorithms
j1, _ = kmedoids.kmedoids(len(numbers),
distance.euclidean_matrix(data2))
j2, _ = kmeans.kmeans(len(numbers), data2, distance.euclidean)
# sklearn k-means
km = KMeans(len(numbers))
j3 = km.fit(data2).labels_
a = metric_func(labels, j1)
b = metric_func(labels, j2)
c = metric_func(labels, j3)
m1.append(a)
m2.append(b)
m3.append(c)
kmedoids_metric.append([np.mean(m1), np.min(m1), np.max(m1)])
kmeans_metric.append([np.mean(m2), np.min(m2), np.max(m2)])
kmeans_sklearn_metric.append([np.mean(m3), np.min(m3), np.max(m3)])
return kmedoids_metric, kmeans_metric, kmeans_sklearn_metric
def gauss_eval(dist_matrix_kmedoids, dist_func_kmeans, metric_func,
nrange=range(10,100,10), num_avg=5):
"""Return metric evaluation on gaussian dataset against N.
Compare K-medoids and K-means.
Input
-----
dist_matrix_kmedoids - function to generate the distance matrix for
kmedoids
dist_func_kmeans - distance function to be used in kmeans
metric_func - metric function being evaluated
nrange - range of N's to be tested, number of data points
num_avg - number of times we cluster the same points and take
the average, min, and max
Output
------
kmedoids_metric - metric computed with K-medoids
kmeans_metric - metric computed with K-means
kmeans_sklearn_metric - metric with kmeans from sklearn
"""
kmedoids_metric = []
kmeans_metric = []
kmeans_sklearn_metric = []
# we generate data with n points in each cluster and evaluate
# the algorithms
for n in nrange:
data = np.concatenate((
np.random.multivariate_normal([0, 0], [[4,0], [0,1]], n),
np.random.multivariate_normal([3, 5], [[1,0.8], [0.8,2]], n),
np.random.multivariate_normal([-2, 3], [[0.5,0], [0,0.5]], n))
)
labels = np.concatenate([[m]*n for m in range(3)])
m1 = []
m2 = []
m3 = []
k = 3
for i in range(num_avg):
j1, _ = kmedoids.kmedoids(k, dist_matrix_kmedoids(data))
j2, _ = kmeans.kmeans(k, data, dist_func_kmeans)
km = KMeans(k)
r = km.fit(data)
j3 = r.labels_
a = metric_func(labels, j1)
b = metric_func(labels, j2)
c = metric_func(labels, j3)
m1.append(a)
m2.append(b)
m3.append(c)
kmedoids_metric.append([np.mean(m1), np.min(m1), np.max(m1)])
kmeans_metric.append([np.mean(m2), np.min(m2), np.max(m2)])
kmeans_sklearn_metric.append([np.mean(m3), np.min(m3), np.max(m3)])
return kmedoids_metric, kmeans_metric, kmeans_sklearn_metric
def dominant_colors(path, k):
with timer("Image loaded: {}"):
image = Image.open(path)
with timer("Transform points: {}"):
pts = points(image)
with timer("Calculate centers: {}"):
centers = kmeans(pts, k)
return centers
def get_dominant_colors(self):
img = Image.open(self.wallpaper)
img.thumbnail((300, 300)) # Resize to speed up python loop.
width, height = img.size
points = self._get_points_from_image(img)
rgbs = kmeans.kmeans(points, self.k)
#rgbs = [map(int, c.center.coords) for c in clusters]
return [self.rgb_to_hex(rgb) for rgb in rgbs]
def test_single_point():
value = [0, 10, 20]
points = [
[value, 1]
]
k = 1
means = kmeans(points, k)
assert 1 == len(means)
assert value == means[0]
def GMM(K, data, stop_times):
num = data.shape[0]
dim = data.shape[1]
clusters = kmeans.kmeans(K, data, 100)
mul = np.zeros((K, dim))
cov = np.zeros((K, dim, dim))
for i in range(K):
mul[i] = np.mean(clusters[i],axis=0)
cov[i] = np.cov(clusters[i].T)
#latent variable
z = np.zeros((num, K))
times = 0
# print data[999]
while(times<stop_times):
#compute
p = np.zeros((num, K))
#E step
for i in range(num):
for j in range(K):
p[i, j] = np.exp(-1/2*(data[i]-mul[j]).dot(inv(cov[j])).dot((data[i]-mul[j]).T))*det(cov[j])**(-1/2)
# print p[i,j]
for j in range(K):
z[i,j] = p[i, j]/np.sum(p[i, :])
#M step
for j in range(K):
tmp = np.zeros((1,dim))
for i in range(num):
tmp += z[i,j]*data[i]
mul[j] = tmp/np.sum(z[:,j])
for j in range(K):
tmp = np.zeros((dim,dim))
for i in range(num):
tmp += z[i,j]*np.outer((data[i]-mul[j]),(data[i]-mul[j]))
cov[j] = tmp/np.sum(z[:,j])
times += 1
# print times
print mul, cov
out = []
fig = plt.figure()
ax = fig.gca(projection='3d')
for i in range(K):
out.append(np.diag(np.exp(-1/2*(data-mul[i]).dot(inv(cov[i])).dot((data-mul[i]).T))/((2*np.pi)*det(cov[i])**(1/2))))
ax.plot_trisurf(data[:,0], data[:,1], out[i], color=np.random.rand(50))
ax.set_xlabel('X')
# ax.set_xlim(-40, 40)
ax.set_ylabel('Y')
# ax.set_ylim(-40, 40)
ax.set_zlabel('Z')
# ax.set_zlim(-100, 100)
plt.show()
def do_analysis(self):
partnum = len(self.cfg.user_subsets)
start_description = range(partnum)
log.info("Performing subset splitting using kmeans")
# Create the first scheme
start_scheme = scheme.create_scheme(self.cfg, "start_scheme", start_description)
for i in start_scheme:
# Save the alignment path
i.make_alignment(self.cfg, self.alignment)
phylip_file = i.alignment_path
print phylip_file
# Add option to output likelihoods, *raxml version takes more
# modfying of the commands in the analyse function
phyml.analyse("GTR", str(phylip_file), "./analysis/start_tree/filtered_source.phy_phyml_tree.txt", "unlinked", "--print_site_lnl")
phyml_lk_file = str(phylip_file) + "_phyml_lk_GTR.txt"
likelihood_dictionary = kmeans.phyml_likelihood_parser(phyml_lk_file)
kmeans.kmeans(likelihood_dictionary)
print start_scheme
def runBenchmark(N=200, M=10000, K=3, usePCA=False, n_components=2):
# note that order of clusters will vary from run to run, so track which true pop with each cluster
# along with true frac count
# sample N indivs without replacement, and also get first M snps for each geno
indices = random.sample(range(len(genoArr_)), N)
indivs_copy = np.array([copy.deepcopy(indivs_[i]) for i in indices])
for i in range(N):
indivs_copy[i].geno = np.array(indivs_copy[i].geno[:M])
indivs_copy[i].j = i # also update position in new indiv list
genoArr_copy = np.array([genoArr_[i][:M] for i in indices])
### TIMING ###
pcaTime = 0
def pca_i(): # zero input fxn for timeit
return PCA_nocluster.pca_transform(indivs_copy, genoArr_copy, n_components)
if usePCA:
#print("timing pca...") # test 1 runs. genoArr_copy isn't changed from run-to-run
#(indivs_copy does get changed, but shouldn't affect run since it restarts each time)
# unlike kmeans, pca is deterministic so runtime shouldn't vary. also pca step is slower, bottleneck.
genoArr_copy_geno = genoArr_copy # make a copy of genotype data first (as opposed to components)
pcaTime = timeit.timeit(pca_i, number = 2)/2.0
# get components for genoArr_copy for kmeans (both timing and quality runs)
# pca deterministic, so don't need to rerun for each quality trial
pcaObj, genoArr_copy = PCA_nocluster.pca_transform(indivs_copy, genoArr_copy, n_components)
def kmeans_i(): # zero input fxn for timeit
return kmeans.kmeans(indivs_copy, genoArr_copy, K)
# timing kmeans, run 10x per data pt, avg
#print("timing k-means 10x...")
kmeansTime = timeit.timeit(kmeans_i, number = 10)/10.0
### QUALITY ###
majFracAvgsByRun = np.zeros(10)
for run in range(10):
#a = np.asarray_chkfinite(indivs_copy)
#a = np.asarray_chkfinite(genoArr_copy)
centers = kmeans.kmeans(indivs_copy, genoArr_copy, K, maxIter = 1000, verbose = False)
kmeansObj = kmeans.kmeansObj(indivs_copy, centers)
majPops, majFracs, clusterSizes = majorityPop(indivs_copy, K)
majFracAvgsByRun[run] = 1.0*sum(majFracs)/K # unweighted avg across clusters
majFrac_avg = np.mean(majFracAvgsByRun)
majFrac_std = np.std(majFracAvgsByRun)
# pcaTime is 0 if pca isn't used
return kmeansObj, majFrac_avg, majFrac_std, kmeansTime, majPops, majFracs, pcaTime
def clusterConvergence2ModesL(xk, ym, Rk, yk):
Np = xk.shape[1]
p = xk.shape[0]
d = ym.shape[0]
# evaluate the unimodal fit
# compute the mean
mu1 = np.mean(xk, axis=1)
# compute the covariance
coef = 1.0 / (float(Np) - 1.0)
Pxx = np.zeros((2, 2))
for k in range(Np):
Pxx = Pxx + coef * np.outer(xk[:, k] - mu1, xk[:, k] - mu1)
Ly1 = np.zeros(Np)
for k in range(Np):
yexp = yk[:, k]
# compute the PDF of y given xk[:,k]
pyx = gaussianNormalPdf(ym - yexp, np.zeros(d), Rk)
# compute the PDF of xk[:,k]
px = gaussianNormalPdf(xk[:, k], mu1, Pxx)
Ly1[k] = pyx * px
# evaluate the bimodal fit
Ly2 = np.zeros(Np)
Pxx2 = np.zeros((2, 2, 2))
mux2 = np.zeros((2, 2))
(idxk, mui) = kmeans.kmeans(xk.transpose(), 2)
for jk in range(2):
idx = np.nonzero(idxk == jk)
idx = idx[0]
# compute the covariance for the jkth mode
N2 = len(idx)
# error checking to prevent single-particle clusters, which don't make sense and break the covariance computation
if N2 == 1:
# set Ly2 to zero & break
Ly2 = np.zeros(Np)
break
coef = 1.0 / (float(N2) - 1.0)
mu2 = np.mean(xk[:, idx], axis=1)
mux2[jk, :] = mu2
Px2 = np.zeros((2, 2))
for k in idx:
Px2 = Px2 + coef * np.outer(xk[:, k] - mu2, xk[:, k] - mu2)
Pxx2[jk, :, :] = Px2.copy()
for k in idx:
yexp = yk[:, k]
# compute the PDF of y given xk[:,k]
pyx = gaussianNormalPdf(ym - yexp, np.zeros(d), Rk)
# compute the PDF of xk[:,k]
px1 = gaussianNormalPdf(xk[:, k], mu2, Px2)
Ly2[k] = pyx * px
print("L1 = %g, L2 = %g" % (Ly1.max(), Ly2.max()))
if not (Ly2.max() > Ly1.max()):
idxk = np.zeros(Np)
# mui = np.mean(xk,axis=1).transpose()
return (1, idxk, mu1, Pxx)
return (2, idxk, mux2, Pxx2)
"""
def test_two_points():
real_mean = [10, 10, 10]
points = [
[(0, 0, 0), 1],
[(20, 20, 20), 1]
]
k = 1
means = kmeans(points, k)
assert 1 == len(means)
assert real_mean == means[0]
def test_two_points_with_weights():
real_mean = [20, 20, 20]
points = [
[(0, 0, 0), 1],
[(30, 30, 30), 2]
]
k = 1
means = kmeans(points, k)
assert 1 == len(means)
assert real_mean == means[0]
def KmeansCluster(FileName = "",Num = 3):
dataSet = []
fileIn = open(FileName)
for line in fileIn.readlines():
lineArr = line.strip().split(',')
dataSet.append([float(i) for i in lineArr])
dataSet = mat(dataSet)
k = Num
centroids, clusterAssment = km.kmeans(dataSet, k)
return centroids,clusterAssment
def clustering_map(cityMap, k):
"""
cluster the city map into k clusters.
"""
# convert all the cityMap vertices into k-means structure
points = [kmeans.Point([cityMap.pos[v][0], cityMap.pos[v][1]], v) for v in cityMap.rv]
# Cluster those data!
opt_cutoff = 0.5
clusters = kmeans.kmeans(points, k, opt_cutoff)
cityMap.node_clusters(clusters)
def test_two_points_two_centers():
values = [
[0, 10, 20],
[-100, -400, -1600]
]
points = [
[value, 1] for value in values
]
k = 2
means = kmeans(points, k)
assert 2 == len(means)
for value in values:
assert value in means
