def _run_interface(self, runtime):
fname = self.inputs.volume
#load data, read lines 8~penultimate
datafile = open(fname, 'rb')
data = [i.strip().split() for i in datafile.readlines()]
stringmatrix = data[8:-1]
datafile.close()
if self.inputs.hemi == 'lh': chosenvertices = lhvertices
if self.inputs.hemi == 'rh': chosenvertices = rhvertices
corrmatrix = np.zeros((len(chosenvertices),len(chosenvertices)))
for x, vertex in enumerate(chosenvertices):
for i in xrange(len(chosenvertices)):
corrmatrix[x][i] = abs(float(stringmatrix[vertex][i]))
if self.inputs.cluster_type == 'spectral':
labels = spectral(corrmatrix, n_clusters=self.inputs.n_clusters, mode='arpack')
if self.inputs.cluster_type == 'hiercluster':
labels = Ward(n_clusters=self.inputs.n_clusters).fit_predict(corrmatrix)
if self.inputs.cluster_type == 'kmeans':
labels = km(n_clusters=self.inputs.n_clusters).fit_predict(corrmatrix)
if self.inputs.cluster_type == 'dbscan':
labels = DBSCAN(eps=np.average(corrmatrix)+np.std(corrmatrix)).fit_predict(corrmatrix)
sxfmout = self.inputs.sxfmout
img = nb.load(sxfmout)
outarray = -np.ones(shape=img.shape[0])
for j, cluster in enumerate(labels):
outarray[chosenvertices[j]] = cluster+1
new_img = nb.Nifti1Image(outarray, img.get_affine(), img.get_header())
_, base, _ = split_filename(fname)
nb.save(new_img, os.path.abspath(base + '_clustered.nii'))
return runtime
def _run_interface(self, runtime):
#load data
data = nb.load(self.inputs.in_File).get_data()
corrmatrix = np.squeeze(data)
if self.inputs.cluster_type == 'spectral':
positivecorrs = np.where(
corrmatrix > 0, corrmatrix,
0) #threshold at 0 (spectral uses non-negative values)
newmatrix = np.asarray(
positivecorrs,
dtype=np.double) #spectral expects dtype=double values
labels = spectral(newmatrix,
n_clusters=self.inputs.n_clusters,
eigen_solver='arpack',
assign_labels='discretize')
if self.inputs.cluster_type == 'hiercluster':
labels = Ward(
n_clusters=self.inputs.n_clusters).fit_predict(corrmatrix)
if self.inputs.cluster_type == 'kmeans':
labels = km(
n_clusters=self.inputs.n_clusters).fit_predict(corrmatrix)
if self.inputs.cluster_type == 'dbscan':
labels = DBSCAN(eps=self.inputs.epsilon).fit_predict(corrmatrix)
new_img = nb.Nifti1Image(labels + 1,
None) #+1 because cluster labels start at 0
_, base, _ = split_filename(self.inputs.in_File)
nb.save(
new_img,
os.path.abspath(base + '_' + str(self.inputs.n_clusters) + '_' +
self.inputs.cluster_type + '_' + self.inputs.hemi +
'.nii'))
return runtime
def run(self, src):
x = self.table(src)
k = self.settings.k
if k == "auto":
raise RuntimeError("Have to implement auto parser")
cl = km(n_clusters=k)
y = cl.fit_predict(x)
clusters = {}
for x_i, y_i in zip(x, y):
clusters[y_i] = clusters.get(y_i, []) + [x_i]
return clusters
def retrain(clusters):
pickle_file = open("train_data.pickle", "rb")
arr = pickle.load(pickle_file)
pickle_file.close()
clt = km(n_clusters = clusters)
clt.fit(arr)
pickle_file = open("clt.pickle", "wb")
pickle.dump(clt, pickle_file)
pickle_file.close()
def __nk(self, nk=None):
if self.Xm is None:
self.nk = None
elif nk is None:
#self.nk = np.random.randint(0,self.k, self.n)
self.nk = km(n_clusters=self.k).fit_predict(self.Xm)
else:
try:
self.nk = nk
self.nk.shape = (self.n, )
except ValueError:
print "nk must be of the same lenght as Xm"
return self.nk
def cluster(nc):
global list_u
global dict_u
clist = []
for i in range(len(list_u)):
clist.append(dict_u[list_u[i]])
result = km(n_clusters=nc,
max_iter=300, n_init=40, init='k-means++').fit_predict(
np.array(clist)) #根据大学的不同学科情况进行聚类，聚类个数为nc个，最大迭代次数为300
ny = [[] for i in range(nc)] #初始化数组
for i in range(top_k):
ny[result[i]].append(list_u[i]) #按照类别将大学名称加入各类的list中
for i in range(nc): #输出各类的大学名称
print(ny[i])
def genlist():
print("fetching candidate replacements for " + self.pred)
cdatasource = eval(self.classname).pred_candidates
if not cdatasource.has(self.lnoun):
print("can't find typical preds for ", self.noun)
return None
cands_max = cdatasource.get(self.lnoun).most_common()
cands = set()
syns = self.get_syns()
syns.sort(key=lambda x: self.modifies_noun(x), reverse=True)
strong_syns = list()
mat = list()
good = list()
for cand in cands_max:
c = cand[0]
if c in vecs and abst.has(c) and abst.get(c) > abst.get(
self.pred):
if c != self.pred and c in syns:
strong_syns.append(c)
mat.append(vecs.get(c))
good.append(cand)
if len(strong_syns) > 0:
self.strong_syns = strong_syns
mat = np.array(mat)
if len(cands_max) > 100 and False:
k = km(n_clusters=round(len(cands_max) / 50),
random_state=0).fit(mat)
hotclust = k.predict(vecs.get(self.pred).reshape(1, -1))
cands = [
cand for cand in good
if k.predict(vecs.get(cand[0]).reshape(1, -1)) == hotclust
]
cands.sort(key=lambda x: x[1])
#cands = squeeze(cands,15)
else:
cands = good
cands = [cand[0] for cand in cands[:50]]
if len(strong_syns) > 1:
cands = set(cands[:5]).union(set(strong_syns))
cands.discard(self.pred)
ret = list(cands)
else:
cands = set(cands[:15]).union(set(syns[:5]))
cands.discard(self.pred)
ret = squeeze(list(cands), 5)
return ret
def input_generator(file_names, n_clust, n_sub, cluster_method, sub_length):
'''
Creates a matrix from the gaze data. Gaze data from each file is clustered to n_clust points using either HR
clustering or KM clustering. These n_clust points are flattened to 2 * n_clust values and each of these 2 * n_clust
values are divided in n_sub subsequences of length sub_length. The columns of the matrix correspond to these gaze
subsequences of length sub_length flattened into 2 * sub_length values.
:param file_names: Gaze files names as the list of strings.
:param n_clust: Number of clusters as an integer.
:param n_sub: Number of Sub seqeunces per image as an integer
:param cluster_method: Clustering method used as a String. Can be either 'HR' or 'KM'
:param sub_length: Integer length of the sub sequence
:return: Input matrix with all the clustered gaze points organized into sub sequences.
'''
mat = []
for image in file_names:
gaze = []
with open(args.gaze_path + image, 'r') as f:
reader = csv.reader(f, delimiter=',')
count = 0
for row in reader:
gaze.append((int(row[0]), int(row[1]), count * 110))
count += 1
gaze = np.array(gaze)
try:
if cluster_method == 'HR':
cluster_labels = hc(n_clusters=n_clust).fit_predict(gaze)
elif cluster_method == 'KM':
cluster_labels = km(n_clusters=n_clust).fit_predict(gaze)
except:
raise ValueError(
'Choose between "HR" or "KM" as the clustering method')
result = {
i: gaze[np.where(cluster_labels == i)]
for i in range(n_clust)
}
centres = []
for cluster in result:
cluster_points = np.array(result[cluster])
cluster_centre = np.mean(cluster_points, axis=0)
centres.append(int(cluster_centre[0]))
centres.append(int(cluster_centre[1]))
# for i in range(0, len(centres) - (2 * sub_length), 2 * ((n_clust - sub_length)/ n_sub)):
for i in range(0, len(centres), ((2 * n_clust) / n_sub)):
mat.append(centres[i:i + (2 * sub_length)])
return np.transpose(np.array(mat))
def _run_interface(self, runtime):
#load data
data = nb.load(self.inputs.in_File).get_data()
corrmatrix = np.squeeze(data)
if self.inputs.cluster_type == 'spectral':
positivecorrs = np.where(corrmatrix>0,corrmatrix,0) #threshold at 0 (spectral uses non-negative values)
newmatrix = np.asarray(positivecorrs,dtype=np.double) #spectral expects dtype=double values
labels = spectral(newmatrix, n_clusters=self.inputs.n_clusters, eigen_solver='arpack', assign_labels='discretize')
if self.inputs.cluster_type == 'hiercluster':
labels = Ward(n_clusters=self.inputs.n_clusters).fit_predict(corrmatrix)
if self.inputs.cluster_type == 'kmeans':
labels = km(n_clusters=self.inputs.n_clusters).fit_predict(corrmatrix)
if self.inputs.cluster_type == 'dbscan':
labels = DBSCAN(eps=self.inputs.epsilon).fit_predict(corrmatrix)
new_img = nb.Nifti1Image(labels+1, None) #+1 because cluster labels start at 0
_, base, _ = split_filename(self.inputs.in_File)
nb.save(new_img, os.path.abspath(base+'_'+str(self.inputs.n_clusters)+'_'+self.inputs.cluster_type+'_'+self.inputs.hemi+'.nii'))
return runtime
def genlist():
print("fetching candidate replacements for "+self.pred)
cdatasource = eval(self.classname).pred_candidates
if not cdatasource.has(self.lnoun):
print("can't find typical preds for ",self.noun)
return None
cands_max = cdatasource.get(self.lnoun).most_common()
cands = set()
syns = self.get_syns()
syns.sort(key=lambda x: self.modifies_noun(x),reverse=True)
strong_syns = list()
mat = list()
good = list()
for cand in cands_max:
c = cand[0]
if c in vecs and abst.has(c) and abst.get(c) > abst.get(self.pred):
if c != self.pred and c in syns:
strong_syns.append(c)
mat.append(vecs.get(c))
good.append(cand)
if len(strong_syns) > 0:
self.strong_syns = strong_syns
mat = np.array(mat)
if len(cands_max) > 100 and False:
k = km(n_clusters=round(len(cands_max)/50),random_state=0).fit(mat)
hotclust = k.predict(vecs.get(self.pred).reshape(1,-1))
cands = [cand for cand in good if k.predict(vecs.get(cand[0]).reshape(1,-1)) == hotclust]
cands.sort(key=lambda x : x[1])
#cands = squeeze(cands,15)
else:
cands = good
cands = [cand[0] for cand in cands[:50]]
if len(strong_syns) > 1:
cands = set(cands[:5]).union(set(strong_syns))
cands.discard(self.pred)
ret = list(cands)
else:
cands = set(cands[:15]).union(set(syns[:5]))
cands.discard(self.pred)
ret = squeeze(list(cands),5)
return ret
def fit(self, original_image, segmented_image):
self.original_image = original_image
self.segmented_image = segmented_image
self.groups = np.unique(
self.segmented_image[self.segmented_image != -1])
self.flat_segments = np.reshape(self.segmented_image, (-1))
self.flat_image = np.reshape(self.original_image, (-1, 3))
# Extracting objects from image
objects = []
for group in self.groups:
object = self.flat_image[self.flat_segments == group]
objects.append(object)
# Constructing features from objects
features = np.zeros((len(objects), 3))
for i, object in enumerate(objects):
max = np.max(object)
min = np.min(object)
mean = np.mean(object)
features[i] = [max, min, mean]
self.group_counts = np.array(self.n_clusters)
kmeans = km(n_clusters=self.n_clusters, random_state=0).fit(features)
labels = kmeans.labels_
labelled_image = np.zeros(self.flat_segments.shape[0])
for i, group in enumerate(self.groups):
labelled_image[self.flat_segments == group] = labels[i]
labelled_image = np.reshape(labelled_image, self.segmented_image.shape)
if self.visualise:
cv.imshow("Clustered image", labelled_image)
cv.waitKey(0)
cv.destroyWindow("Clustered image")
return labelled_image
def kmeans(
feature_matrix: pd.DataFrame(), k: int = 2,
feature_columns: list = []):
"""包裝Kmeans
Args:
feature_matrix (pd.DataFrame): 經過前處理的特徵矩陣
k (int, optional): 分K群. Defaults to 2.
feature_columns (list, optional): 選擇分群用的特徵的columns. Defaults to [].
Returns:
sklearn.cluster.KMeans: sklearn Kmeans分群結果物件
"""
if not feature_columns:
feature_columns = feature_matrix.columns
# uid以外作為特徵
feature_matrix = feature_matrix[feature_columns]
# 轉np array做輸入用
x = np.array(feature_matrix)
# 跑Kmeans分群
km_model = km(n_clusters=k, random_state=0).fit(x)
return km_model
def silhoutte(name):
df = pd.read_csv("../processing/" + name + "_dist.csv.virus")
xs = list(df['x'])
ys = list(df['y'])
xs = [x - min(xs) for x in xs]
ys = [x - min(ys) for x in ys]
X = np.matrix(zip(xs, ys))
stat = open('kstatistics.csv', 'w')
ncluster = []
distortion_set = []
silh = []
for nc in range(2, 13):
kmeans = km(n_clusters=nc).fit(X)
cluster_labels = kmeans.fit_predict(X)
silhouette_avg = silhouette_score(X, cluster_labels)
distortion = 0
distortion = (sum(
np.min(cdist(X, kmeans.cluster_centers_, 'euclidean'), axis=1)) /
X.shape[0])
ncluster.append(nc)
distortion_set.append(distortion)
silh.append(silhouette_avg)
# print kmeans.cluster_centers_, sum(np.min(cdist(X,
# kmeans.cluster_centers_, 'euclidean'), axis=1))/13
print(nc, distortion, silhouette_avg)
print(nc, distortion, silhouette_avg, file=stat)
fig = plt.figure()
ax = fig.add_subplot(111)
plt.plot(ncluster, distortion_set, '-o')
ax.set_xlabel('No.of clusters')
ax.set_ylabel('Distortion')
ax.set_title("Selecting K with Elbow method", fontsize=10)
plt.show(block=False)
''' fig1 = plt.figure()
a = exp_X_cent_dist / tf.reduce_sum(
exp_X_cent_dist, axis=2, keep_dims=True)
print a
return a
def fit(self, X_train):
c, l = self.sess.run(self.network, {self.X: X_train})
return c, l
from sklearn.cluster import KMeans as km
if __name__ == "__main__":
nb_samples = 10000
E = 2
nb_clusters = 2
X, y = make_blobs(n_samples=nb_samples, centers=nb_clusters, n_features=E)
X_ = X[np.newaxis, :]
y = y[np.newaxis, :]
print y
kmean = KMeans(nb_clusters)
kmean.init()
centroids, labels = kmean.fit(X_)
print centroids
print np.sum(labels)
print y
kmeans = km(n_clusters=2, random_state=0).fit(X)
print kmeans.cluster_centers_
#normalization
def fun(i):
x = ((i - i.min()) / (i.max() - i.min()))
return (x)
df_norm = fun(df.iloc[:, 1:])
df_norm.describe()
# In[157]:
#elbow curve
wss = []
k = list(range(10, 100, 5))
for i in k:
kmeans = km(n_clusters=i)
kmeans.fit(df_norm)
wss.append(kmeans.inertia_)
wss
# In[158]:
plt.plot(k, wss, 'ro-')
plt.xlabel('number of clusters')
plt.ylabel('total with in ss')
# In[159]:
model = km(n_clusters=40)
model.fit(df_norm)
model.labels_
print('Computing using: ' + method + ' breed method')
start = datetime.now()
GENERATION = copy.deepcopy(init_pop)
for i in range(gens):
print('Generation no: ' + str(i + 1))
GENERATION.select()
survivors.append(GENERATION.population)
top_scores.append((min(GENERATION.score())))
fittest.append(GENERATION.population[GENERATION.sorted_scores[0]])
GENERATION.mutate(0.001)
GENERATION.breed(method=method)
GENERATION.population = fittest
GENERATION.sorted_scores = np.argsort(GENERATION.score())
fit_rank = GENERATION.sorted_scores
alpha = fittest[fit_rank[0]]
ga_means = km(n_clusters, alpha, 1).fit(X)
cluster_list.append(ga_means)
end = datetime.now()
comp_duration.append(end - start)
Fittest.append(alpha)
#Survivors.append(survivors)
for i in range(len(cluster_list)):
Distances.append(galuster.sum_distances(cluster_list[i], X))
plt.figure(i)
galuster.lolipop_plot(cluster_list[i], X)
#Iterate GA operations over number of generations
#ga_start = datetime.now()
#
frqs = [['Actual/Predict', 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 'Recall']
] + [[i] + [0 for j in range(10)] for i in range(10)] + [['Precision']]
# saving original pics
# for i in range(2000):
# scipy.misc.imsave("Output/2.2c/Original_"+str(n_cluster)+"/"+str(i)+".bmp",data[i].reshape(28,28))
# modeling and predicting labels
pca = PCA(n_components=.9)
data = pca.fit_transform(data)
print data[0]
sys.exit()
model = km(n_clusters=n_cluster,
max_iter=2000,
n_init=100,
init='k-means++',
tol=.00001,
n_jobs=4)
model.fit(np.array(data))
print 1
p_label = model.fit_predict(data)
print 2
# finding mapping
for i in range(2000):
result[p_label[i]].append(labels[i][0])
mapping = [max(set(i), key=i.count) for i in result]
# saving pics after applying PCA
for i in range(2000):
scipy.misc.imsave(
def find_spread(X,cluster_size):
X_kmeans = km(cluster_size, random_state = 0).fit(X)
X_kmeans_list = {i: X[np.where(X_kmeans.labels_ == i)] for i in range(cluster_size)}
X_spread = list(np.diag(np.diag(np.cov(X_kmeans_list[i].T))) for i in range(cluster_size))
return X_spread
ga_start = datetime.now()
for i in range(gens):
print('Generation no: ' + str(i + 1))
pop.select()
#survivors.append(pop.population)
top_scores.append((min(pop.score())))
fittest.append(pop.population[pop.sorted_scores[0]])
pop.mutate(0.001)
pop.breed(method='hybrid')
#Cluster the data using the fittest seed
init_pop.population = fittest
init_pop.sorted_scores = np.argsort(init_pop.score())
fit_rank = init_pop.sorted_scores
ga_means = km(n_clusters, fittest[fit_rank[0]], 1).fit(X)
ga_end = datetime.now()
comp_duration.append(ga_end - ga_start)
cluster_list.append(ga_means)
sum_of_dist.append(galuster.sum_distances(ga_means, X))
#Plot GA cluster membership
plt.figure(0)
galuster.lolipop_plot(ga_means, X)
for i in range(no_kmeans):
print('Starting kmeans algorithm no: ' + str(i + 1))
start = datetime.now() #Record starting time
kmeans = km(n_clusters, n_init=n_seed).fit(X) #compute kmeans
return clusters
def organizeEnumeratedDictionary(dictionary):
newDict = {}
for i in range(0, len(dictionary)):
newDict[i]=dictionary[i]
return newDict
def emojiCodeToEmoji(clusterDict, emojiDict):
for i in clusterDict:
for x in range(0, len(clusterDict[i])):
clusterDict[i][x] = emojiDict[clusterDict[i][x]]
return clusterDict
tsvToEmojiDict = arrayToDict(getFile('emoji_lookup.tsv'))#dictionary to translate emojis over
emojiDataFrame = pd.read_csv(StringIO(codecs.open('emojis.txt', 'r', encoding='utf8', errors='ignore').read()), sep='\s+')#creates pandas system for holding data
dimensions = np.array(emojiDataFrame.as_matrix(columns=emojiDataFrame.columns[1:]))#gets dimensions in an array perfect for cluster
# auto random state
cluster = km(n_clusters = 100, max_iter=10000000)#kmeans clustering
#cluster = ap(max_iter=1000000)#Affinity propogation clustering
#cluster = b(n_clusters=200)#birch clustering
cluster.fit(dimensions)
codedCluster = addEmojiCode(dimensions, cluster, getFile('emoji_lookup.tsv'))
organizedClusters = organizeEnumeratedDictionary(codedCluster)
emojiClusters = emojiCodeToEmoji(organizedClusters, tsvToEmojiDict)
for z in emojiClusters:
print("Clusters" + str(z))
print(emojiClusters[z])
print("\n")
"""
import pandas as pd
from sklearn.cluster import KMeans as km
import numpy
import matplotlib.pyplot as plt
from sklearn.metrics import silhouette_score as ss
data = pd.read_csv("sdm.csv", sep=";", header=None)
data.columns = ["depth", "param_1", "param_2"]
data_pr = data[["param_1", "param_2"]]
plt.scatter(data_pr.param_1, data_pr.param_2)
kmeans = km(init='k-means++', n_clusters=3, random_state=0).fit(data_pr.as_matrix())
# data_pr['labels'] =pd.Series(kmeans.labels_)
# data_pr.plot.scatter(x='b',y='c',c='labels', colormap='viridis')
data_pr1 = data_pr.copy()
scores = []
for k in range(2, 10):
kmeans = km(init='k-means++', n_clusters=k, random_state=0).fit(data_pr.as_matrix())
data_pr1['labels'] = pd.Series(kmeans.labels_)
print(len(kmeans.labels_))
data_pr1.plot.scatter(x='b', y='c', c='labels', colormap='viridis')
scores.append(ss(data_pr1[['b', 'c']], labels=data_pr1['labels']))
print(data_pr1)
n = [i for i in range(2, 10)]
plt.figure()
def run(self, clusters=3):
self.y = km(n_clusters=clusters).fit_predict(self.X)
data.head()
#plotting values
f1 = data['duration'].values
f2 = data['imdb_score'].values
X = np.array(list(zip(f1, f2)))
print("This is X. The zipped array")
print(X)
#Finding optimal k
wcss = []
for i in range(1, 11):
kmeans = km(n_clusters=i,
init='k-means++',
max_iter=300,
n_init=10,
random_state=0)
kmeans.fit(X)
wcss.append(kmeans.inertia_)
#Plotting the results onto a line graph, allowing us to observe 'The elbow'
plt.plot(range(1, 11), wcss)
plt.title('The elbow method')
plt.xlabel('Number of clusters')
plt.ylabel('WCSS') #within cluster sum of squares
plt.show()
k = 4
kmeans = km(n_clusters=k)
KMmodel = kmeans.fit(X)
for line in lines:
if line.strip() == "":
break
name = line.split(":")[0].strip()
location = line.split(":")[1].strip().split(",")
top_yelp.append([name, float(location[0].strip()), float(location[1].strip())])
top_yelp_lat_long = []
for row in top_yelp:
top_yelp_lat_long.append([row[1], row[2]])
t = 21
lat_long = []
for row in all_data[t]:
lat_long.append([row[3], row[4]])
for k in [240]:
kmeans = km(k, max_iter=1000, n_init=50, init="k-means++")
kmeans.fit(lat_long)
pred = kmeans.predict(top_yelp_lat_long)
pred_cluster_centers = [kmeans.cluster_centers_[i] for i in pred]
error = [vincenty(top_yelp_lat_long[i], pred_cluster_centers[i]).miles for i in range(len(top_yelp_lat_long))]
print "For t = {t_val}:".format(t_val=t)
print "Min Error\t\tMax Error\tAvg Error".format()
print "{min_e}\t{max_e}\t{avg_e}".format(min_e=min(error), max_e=max(error), avg_e=sum(error) / len(error))
def runROIDetector(p, k, min_dist, threshold, gammae, nue):
k = k.astype(np.int64)
p = p.astype(np.int64)
threshold = threshold.astype(np.int64)
resize = .3
filesHtrain = []
filesUtrain = []
filesUtest = []
filesHtest = []
filesUCV = []
filesHCV = []
unhealthyTestPatient = [
f for f in listdir('testROI/unhealthy')
if isdir(join('testROI/unhealthy', f))
]
healthyTestPatient = [
f for f in listdir('testROI/healthy')
if isdir(join('testROI/healthy', f))
]
unhealthyTrainPatient = [
f for f in listdir('trainROI/unhealthy')
if isdir(join('trainROI/unhealthy', f))
]
healthyTrainPatient = [
f for f in listdir('trainROI/healthy')
if isdir(join('trainROI/healthy', f))
]
unhealthyCVPatient = [
f for f in listdir('trainROI/unlabel')
if isdir(join('trainROI/unlabel', f))
]
healthyCVPatient = healthyTrainPatient[0:8]
healthyTrainPatient = healthyTrainPatient[8:len(healthyTrainPatient) - 1]
for i in healthyTrainPatient:
dirs = listdir('trainROI/healthy/' + i)
for j in dirs:
if "._" not in j:
filesHtrain.append(('trainROI/healthy/' + i + '/' + j))
for i in unhealthyTrainPatient:
dirs = listdir('trainROI/unhealthy/' + i)
for j in dirs:
if "._" not in j:
filesUtrain.append(('trainROI/unhealthy/' + i + '/' + j))
for i in unhealthyCVPatient:
dirs = listdir('trainROI/unlabel/' + i)
for j in dirs:
if "._" not in j:
filesUCV.append(('trainROI/unlabel/' + i + '/' + j))
for i in healthyTestPatient:
dirs = listdir('testROI/healthy/' + i)
for j in dirs:
if "._" not in j:
filesHtest.append(('testROI/healthy/' + i + '/' + j))
for i in unhealthyTestPatient:
dirs = listdir('testROI/unhealthy/' + i)
for j in dirs:
if "._" not in j:
filesUtest.append(('testROI/unhealthy/' + i + '/' + j))
init_words = []
words = []
final_words = []
healthyPatientDict = {}
# Part One:----------Extract initial words from training positive set----------
for image in (filesUtrain):
# read image from Unhealthy class (These are the seed examples)
img = imread(image)
[h, w, d] = img.shape
img = rescale(img, resize)
[h, w, d] = img.shape
# Trim the image to a bucketable size
F = img[0:int(p * math.floor(h / p)), 0:int(p * math.floor(w / p)), :]
# Extract HOG Features
fd = hog(F[:, :, 1],
orientations=8,
pixels_per_cell=(p, p),
cells_per_block=(1, 1))
# Reorganize features so that each index matches a sample
features = np.reshape(fd, (((F.shape[1] * F.shape[0]) / (p * p)), 8))
# Cluster the features
KM = km(n_clusters=k).fit(features)
# Add the cluster centers to the inital dictionary
init_words.extend(KM.cluster_centers_)
init_words = np.asarray(init_words)
occurencVec = {}
### Part Two:----------------------Filter out "bad words"------------------------
for patient in (healthyTrainPatient):
healthyPatientDict[patient] = {}
for idx in range(len(init_words)):
temp = healthyPatientDict[patient]
temp[idx] = -1
occurencVec[idx] = 0
healthyPatientDict[patient] = temp
for patient in (healthyTrainPatient):
imageList = listdir('trainROI/healthy/' + patient)
HOGvec = []
count = 0
for image in imageList:
if "._" not in image:
# read image from Healthy class
img = imread('trainROI/healthy/' + patient + '/' + image)
[h, w, d] = img.shape
img = rescale(img, resize, anti_aliasing=True)
[h, w, d] = img.shape
# Trim the image to a bucketable size
F = img[0:int(p * math.floor(h / p)),
0:int(p * math.floor(w / p)), :]
# Extract HOG Features
fd = hog(F[:, :, 1],
orientations=8,
pixels_per_cell=(p, p),
cells_per_block=(1, 1))
# Reorganize features so that each index matches a sample
features = np.reshape(fd, (((F.shape[1] * F.shape[0]) /
(p * p)), 8))
if count == 0:
HOGvec = features
else:
HOGvec = np.vstack((HOGvec, features))
count = count + 1
# For each of the words in the inital dictionary, calculate the
# L2-distance between the features of the current image. If the distance
# is too small too many times, remove the word from the dictionary.
initWordsIdx = 0
for rows in (init_words):
num_of_matches = 0
iters = 0
for n in HOGvec:
iters = iters + 1
r = np.linalg.norm(rows - n)
if r < min_dist:
num_of_matches = num_of_matches + 1
temp = occurencVec[initWordsIdx]
occurencVec[initWordsIdx] = temp + 1
iters = iters + 1
final_words.append(rows)
temp = healthyPatientDict[patient]
temp[initWordsIdx] = num_of_matches
initWordsIdx = initWordsIdx + 1
averages = []
for count in range(len(healthyPatientDict[healthyTrainPatient[0]])):
numerator = 0
for patient in healthyTrainPatient:
temp = healthyPatientDict[patient]
numerator = numerator + temp[count]
average = numerator / len(healthyPatientDict)
averages.append(average)
idxs = np.where(np.asarray(averages) < threshold)
idxs = idxs[0]
featureMatrix = np.zeros(shape=(len(healthyPatientDict), len(idxs)))
i = 0
for patient in healthyPatientDict:
j = 0
for idx in idxs:
temp = healthyPatientDict[patient]
featureMatrix[i, j] = temp[idx]
j = j + 1
i = i + 1
validationMatrix = np.zeros(shape=(len(unhealthyCVPatient) +
len(healthyCVPatient), len(idxs)))
i = 0
for patient in (healthyCVPatient):
imageList = listdir('trainROI/healthy/' + patient)
HOGvec = []
count = 0
j = 0
for image in imageList:
if "._" not in image:
# read image from Healthy class
img = imread('trainROI/healthy/' + patient + '/' + image)
[h, w, d] = img.shape
img = rescale(img, resize, anti_aliasing=True)
[h, w, d] = img.shape
# Trim the image to a bucketable size
F = img[0:int(p * math.floor(h / p)),
0:int(p * math.floor(w / p)), :]
# Extract HOG Features
fd = hog(F[:, :, 1],
orientations=8,
pixels_per_cell=(p, p),
cells_per_block=(1, 1))
# Reorganize features so that each index matches a sample
features = np.reshape(fd, (((F.shape[1] * F.shape[0]) /
(p * p)), 8))
if count == 0:
HOGvec = features
else:
HOGvec = np.vstack((HOGvec, features))
count = count + 1
# For each of the words in the inital dictionary, calculate the
# L2-distance between the features of the current image. If the distance
# is too small too many times, remove the word from the dictionary.
initWordsIdx = 0
for num in range(len(idxs)):
num_of_matches = 0
iters = 0
for n in HOGvec:
iters = iters + 1
r = np.linalg.norm(init_words[num] - n)
if r < min_dist:
num_of_matches = num_of_matches + 1
iters = iters + 1
validationMatrix[i, j] = num_of_matches
j = j + 1
i = i + 1
for patient in (unhealthyCVPatient):
imageList = listdir('trainROI/unlabel/' + patient)
HOGvec = []
count = 0
j = 0
for image in imageList:
if "._" not in image:
# read image from Healthy class
img = imread('trainROI/unlabel/' + patient + '/' + image)
[h, w, d] = img.shape
img = rescale(img, resize, anti_aliasing=True)
[h, w, d] = img.shape
# Trim the image to a bucketable size
F = img[0:int(p * math.floor(h / p)),
0:int(p * math.floor(w / p)), :]
# Extract HOG Features
fd = hog(F[:, :, 1],
orientations=8,
pixels_per_cell=(p, p),
cells_per_block=(1, 1))
# Reorganize features so that each index matches a sample
features = np.reshape(fd, (((F.shape[1] * F.shape[0]) /
(p * p)), 8))
if count == 0:
HOGvec = features
else:
HOGvec = np.vstack((HOGvec, features))
count = count + 1
# For each of the words in the inital dictionary, calculate the
# L2-distance between the features of the current image. If the distance
# is too small too many times, remove the word from the dictionary.
initWordsIdx = 0
for num in range(len(idxs)):
num_of_matches = 0
iters = 0
for n in HOGvec:
iters = iters + 1
r = np.linalg.norm(init_words[num] - n)
if r < min_dist:
num_of_matches = num_of_matches + 1
iters = iters + 1
validationMatrix[i, j] = num_of_matches
j = j + 1
i = i + 1
normalizedXtrain = normalize(featureMatrix)
normalizedXCV = normalize(validationMatrix)
clf = svm.OneClassSVM(nu=nue, kernel="rbf", gamma=gammae)
clf.fit(normalizedXtrain)
y_CV = np.ones(shape=(len(unhealthyCVPatient) + len(healthyCVPatient), 1))
y_CV[len(healthyCVPatient):len(unhealthyCVPatient) +
len(healthyCVPatient)] = -1
y_pred_train = clf.predict(normalizedXtrain)
y_pred_CV = clf.predict(normalizedXCV)
return [y_pred_train, y_pred_CV, y_CV]
</script>
<script async defer
src="https://maps.googleapis.com/maps/api/js?key=AIzaSyDu4tAkj9-8cwEPTamK812YSbPnZ6xq9D8&signed_in=true&libraries=visualization&callback=initMap">
</script>
</body>
</html>
"""
all_data = pickle.load(open("../data/all_data_new.p", "rb"))
t = 0
lat_long = []
for row in all_data[t]:
lat_long.append([row[3], row[4]])
kmeans = km(190, max_iter=1000, n_init = 50,init = 'k-means++')
kmeans.fit(lat_long)
t = 0
lat_lon_values = ""
for c in kmeans.cluster_centers_:
lat_lon_values += "new google.maps.LatLng(" + str(c[0]) + ", " + str(c[1]) + "),\n"
lat_lon_values = lat_lon_values[:-2]
file_ = open('../outputs/Google_Heatmap_{t_val}.html'.format(t_val=t), 'w')
file_.write(html_front + lat_lon_values + html_back)
file_.close()
print("How do I look?")
print(features.head())
# ###Normalizing the data and clustering
cols_to_norm = ['Duration','distance_start_stop', 'day_of_week', 'hours']
features[cols_to_norm] = features[cols_to_norm].apply(lambda x: (x - x.mean()) / (x.max() - x.min()))
print("Normalized")
print(features.head(2))
cluster_num = 11
model = km(n_clusters = cluster_num, n_init=5, max_iter=20)
model.fit_transform(features)
print("Model created")
features['labels'] = model.labels_
print("Got some labels.")
print(features.head(2))
#sampling my data to run the silhouette score
sample = features.sample(4000)
silhouette_score(sample.values, sample['labels'].values)
def choose_center(X):
X_kmeans = km(n_clusters = 10, random_state = 0).fit(X)
X_centers = X_kmeans.cluster_centers_
return X_centers
import pandas as pd
from sklearn.cluster import KMeans as km
import numpy as np
import seaborn as sns
#data okuma
df = pd.read_csv("Final-data.txt")
# k ve model olusturma
k = int(input("k:"))
a = km(n_clusters=k).fit(df)
#Perfomance
TCSS = np.sum((df.values - np.sum(df.values, axis=0) / len(df))**2)
WCSS = np.zeros(k)
for index, i in enumerate(a.labels_):
WCSS[i] += np.sum((df.values[index] - a.cluster_centers_[i])**2)**0.5
BCSS = TCSS - np.sum(WCSS)
dist = []
for i in a.cluster_centers_:
for j in a.cluster_centers_:
dist.append(np.sum((i - j)**2)**0.5)
DunnIndex = np.min(WCSS) / np.max(dist)
#####################################################
#sonuclari yazma
c = np.zeros(k)
f = open("sonuc.txt", "w")
for i in range(len(a.labels_)):
f.write("Kayit " + str(i) + ":\t" + "Kume " + str(a.labels_[i]) + "\n")
c[a.labels_[i]] += 1
b0 = []
b1 = []
for i in range(len(a)):
if i % 2 == 0:
b0.append(a[i])
else:
b1.append(a[i])
a0 = [float(x) for x in b0]
a1 = [float(x) for x in b1]
df = pd.DataFrame()
df['x'] = a0
df['y'] = a1
y_true = [0 for i in range(500)]
for i in range(len(a0) - 500):
y_true.append(1)
model = km(n_clusters=3)
y = model.fit_predict(df)
plt.title("KMeans")
plt.scatter(df[y == 0]['x'], df[y == 0]['y'])
plt.scatter(df[y == 1]['x'], df[y == 1]['y'])
plt.show()
print("Purity score for KMeans: ", purity_score(y_true, y))
clustering = AgglomerativeClustering().fit(df)
y = clustering.labels_
plt.title("Agglo_Clustering")
plt.scatter(df[y == 0]['x'], df[y == 0]['y'])
plt.scatter(df[y == 1]['x'], df[y == 1]['y'])
plt.show()
print("Purity score for Agglomerative Clustering: ", purity_score(y_true, y))
def Prototyping(X, numP):
from sklearn.cluster import KMeans as km
kmeans = km(init='k-means++', n_clusters=numP)
kmeans.fit(X)
centers = kmeans.cluster_centers_
return centers
def scale(array):
minimum = array.min()
maximum = array.max()
scaled_array = np.array([])
for i in range(0, len(array)):
scaled_array = np.append(scaled_array,
(array[i] - minimum) / (maximum - minimum))
return scaled_array
# In[]:
ofc_scaled = scale(df_full['ofc'])
mi_scaled = scale(df_full['transform'])
points = np.column_stack((ofc_scaled, mi_scaled))
#metrics = [df_full['ofc'], df_full['mi']]
#points = pd.concat(metrics, axis = 1)
from sklearn.cluster import KMeans as km
kmeans = km(n_clusters=3)
# fit kmeans object to data
kmeans.fit(points)
# print location of clusters learned by kmeans object
print(kmeans.cluster_centers_)
# save new clusters for chart
y_km = kmeans.fit_predict(points)
#plt.scatter(points[y_km ==0,0], points[y_km == 0,1], s=100, c='red')
#plt.scatter(points[y_km ==1,0], points[y_km == 1,1], s=100, c='black')
from sklearn.cluster import KMeans as km
from sklearn.metrics import silhouette_score
import sys
from scipy.spatial.distance import cdist
import numpy as np
df = pd.read_csv("data_dist.csv.virus")
xs = list(df['x'])
ys = list(df['y'])
xs = xs - min(xs)
ys = ys - min(ys)
X = np.matrix(zip(xs, ys))
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
X = scaler.fit_transform(X)
stat = open('kstatistics.csv', 'w')
for nc in range(2, 13):
kmeans = km(n_clusters=nc, random_state=10)
cluster_labels = kmeans.fit_predict(X)
silhouette_avg = silhouette_score(X, cluster_labels)
#print("For n_clusters =", no_cluster, "The average silhouette_score is :", silhouette_avg)
labels = kmeans.labels_
dist = kmeans.transform(X)
distortion = 0
distortion = (
sum(np.min(cdist(X, kmeans.cluster_centers_, 'euclidean'), axis=1)) / X.shape[0])
# print kmeans.cluster_centers_, sum(np.min(cdist(X, kmeans.cluster_centers_, 'euclidean'), axis=1))/13
print nc, distortion, silhouette_avg
print >>stat, nc, distortion, silhouette_avg
im.show()
sys.exit()
labels = mat["data_labels"]
frqs = [['Actual/Predict',0,1,2,3,4,5,6,7,8,9,'Recall']]+[[i]+[0 for j in range(10)] for i in range(10)]+[['Precision']]
# saving original pics
# for i in range(2000):
# scipy.misc.imsave("Output/2.2c/Original_"+str(n_cluster)+"/"+str(i)+".bmp",data[i].reshape(28,28))
# modeling and predicting labels
pca = PCA(n_components = .9)
data = pca.fit_transform(data)
print data[0]
sys.exit()
model = km(n_clusters=n_cluster, max_iter=2000, n_init=100, init='k-means++', tol=.00001, n_jobs=4)
model.fit(np.array(data))
print 1
p_label = model.fit_predict(data)
print 2
# finding mapping
for i in range(2000):
result[p_label[i]].append(labels[i][0])
mapping = [max(set(i), key=i.count) for i in result]
# saving pics after applying PCA
for i in range(2000):
scipy.misc.imsave("../Output/2.2c/After_PCA_"+str(n_cluster)+"/"+str(i)+".bmp",data[i][:81].reshape(9,9))
# saving results
valores=data.values escal=pre.MinMaxScaler() x_esc=escal.fit_transform(valores) x_normalizado=pd.DataFrame(x_esc) pca=PCA(n_components=2) reduced=pd.DataFrame(pca.fit_transform(x_normalizado)) reduced['x']=reduced[0] reduced['y']=reduced[1] from sklearn.cluster import KMeans as km rede=km(n_clusters=4) rede.fit(reduced) lista=np.array([rede.labels_,nomes.to_numpy()]) reduced['cluster']=rede.labels_.tolist() reduced['nomes']=nomes import matplotlib.pyplot as plt plt.scatter(reduced['x'],reduced['y'],c=reduced['cluster'],s=150) [plt.text(reduced['x'][i],reduced['y'][i], nomes.to_numpy()[i]) for i in range(len(nomes))]
cm.yaxis.set_ticklabels(cm.yaxis.get_ticklabels(), rotation=90)
cm.xaxis.set_ticklabels(cm.xaxis.get_ticklabels(), rotation=0)
plt.ylabel('True label')
plt.xlabel('Predicted label')
def convertCluster2Label(cluster_labels, original_labels, labels2convert):
converted_labels = np.full(labels2convert.size, -1)
for i in np.unique(cluster_labels):
temp_original_labels = original_labels[cluster_labels == i]
temp_label = np.bincount(temp_original_labels).argmax()
converted_labels[labels2convert == i] = temp_label
return converted_labels
iris = datasets.load_iris()
X = iris.data
y = iris.target
target_names = iris.target_names
X_train, X_test, y_train, y_true = train_test_split(X, y)
kmeans = km(n_clusters=3, n_init=2020)
kmeans.fit(X_train)
values = kmeans.cluster_centers_.squeeze()
trained_labels = kmeans.labels_
labels_predict = kmeans.predict(X_test)
print(labels_predict)
y_predict = convertCluster2Label(trained_labels, y_train, labels_predict)
print(y_predict)
confusionM(y_true, y_predict, target_names)
