def read_pic(fn):
"""
read_pic
:param fn:
:return:
"""
fnimg = cv2.imread(fn)
img = cv2.resize(fnimg, (500, 400))
w = img.shape[1]
h = img.shape[0]
w_interval = w / w_fg
h_interval = h / h_fg
alltz = []
for now_h in xrange(0, h, h_interval):
for now_w in xrange(0, w, w_interval):
b = img[now_h:now_h + h_interval, now_w:now_w + w_interval, 0]
g = img[now_h:now_h + h_interval, now_w:now_w + w_interval, 1]
r = img[now_h:now_h + h_interval, now_w:now_w + w_interval, 2]
btz = np.mean(b)
gtz = np.mean(g)
rtz = np.mean(r)
alltz.append([btz, gtz, rtz])
result_alltz = np.array(alltz).T
pca = mlpy.PCA()
pca.learn(result_alltz)
result_alltz = pca.transform(result_alltz, k=len(result_alltz) / 2)
result_alltz = result_alltz.reshape(len(result_alltz))
return result_alltz
def readpic(fn):
#返回图片特征码
fnimg = cv2.imread(fn)
img=cv2.resize(fnimg,(500,400))
w=img.shape[1]
h=img.shape[0]
w_interval=w/20
h_interval=h/10
alltz=[]
for now_h in xrange(0,h,h_interval):
for now_w in xrange(0,w,w_interval):
b = img[now_h:now_h+h_interval,now_w:now_w+w_interval,0]
g = img[now_h:now_h+h_interval,now_w:now_w+w_interval,1]
r = img[now_h:now_h+h_interval,now_w:now_w+w_interval,2]
btz=np.mean(b)
gtz=np.mean(g)
rtz=np.mean(r)
alltz.append([btz,gtz,rtz])
result_alltz=np.array(alltz).T
print result_alltz
pca = mlpy.PCA() #进行PCA 降为提取
pca.learn(result_alltz)
result_alltz = pca.transform(result_alltz, k=len(result_alltz)/2)
result_alltz =result_alltz.reshape(len(result_alltz))
print result_alltz
return result_alltz
def get_pca(X, n): ''' Takes X and convert it into n dimensional space and returns the answer ''' pca = mlpy.PCA() pca.learn(X) return pca.transform(X, k = n)
def get_distance(img, findimg):
newsize = (21, 21)
fimg = cv2.resize(findimg, newsize)
img = cv2.resize(img, newsize)
my_img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
my_fimg = cv2.cvtColor(fimg, cv2.COLOR_BGR2GRAY)
pcaimg = mlpy.PCA()
pcaimg.learn(my_img)
pca_img = pcaimg.transform(my_img, k=1)
pca_img = pcaimg.transform_inv(pca_img)
pcafimg = mlpy.PCA()
pcafimg.learn(my_fimg)
pca_fimg = pcaimg.transform(my_fimg, k=1)
pca_fimg = pcafimg.transform_inv(pca_fimg)
return get_EuclideanDistance(pca_img, pca_fimg)
def metric(self):
totalTimer = Timer()
with totalTimer:
model = mlpy.PCA(**self.build_opts)
model.learn(self.data[0])
out = model.transform(self.data[0], self.k)
metric = {}
metric["runtime"] = totalTimer.ElapsedTime()
return metric
def plot_data(X, y):
pca = mlpy.PCA()
pca.learn(X)
z = pca.transform(X, k=2)
plt.set_cmap(plt.cm.Paired)
fig1 = plt.figure(1)
title = plt.title("PCA on mushroom dataset")
plot = plt.scatter(z[:, 0], z[:, 1], c=y)
labx = plt.xlabel("First component")
laby = plt.ylabel("Second component")
plt.show()
def reduce_PCA(x):
'''
Reduce the dimensions using Principal Component
Analysis
'''
# create the PCA object
pca = ml.PCA(whiten=True)
# learn the principal components from all the features
pca.learn(x)
# return the object
return pca
def pcaDimRed(features, nDims):
[X, Y] = listOfFeatures2Matrix(features)
pca = mlpy.PCA(method='cov')
pca.learn(X)
coeff = pca.coeff()
coeff = coeff[:, 0:nDims]
featuresNew = []
for f in features:
ft = f.copy()
# ft = pca.transform(ft, k=nDims)
ft = numpy.dot(f, coeff)
featuresNew.append(ft)
return (featuresNew, coeff)
def RunPCAMlpy(q):
totalTimer = Timer()
# Load input dataset.
Log.Info("Loading dataset", self.verbose)
data = np.genfromtxt(self.dataset, delimiter=',')
try:
with totalTimer:
# Find out what dimension we want.
if "new_dimensionality" in options:
k = int(options.pop("new_dimensionality"))
if (k > data.shape[1]):
Log.Fatal("New dimensionality (" + str(k) +
") cannot be greater " +
"than existing dimensionality (" +
str(data.shape[1]) + ")!")
q.put(-1)
return -1
else:
k = data.shape[1]
build_opts = {}
if "whiten" in options:
build_opts["whiten"] = True
options.pop("whiten")
if len(options) > 0:
Log.Fatal("Unknown parameters: " + str(options))
raise Exception("unknown parameters")
# Perform PCA.
prep = mlpy.PCA(**build_opts)
prep.learn(data)
out = prep.transform(data, k)
except Exception as e:
Log.Fatal("Exception: " + str(e))
q.put(-1)
return -1
time = totalTimer.ElapsedTime()
q.put(time)
return time
def textListToColors(names): ''' Generates a list of colors based on a list of names (strings). Similar strings correspond to similar colors. ''' # STEP A: compute strings distance between all combnations of strings Dnames = np.zeros( (len(names), len(names)) ) for i in range(len(names)): for j in range(len(names)): Dnames[i,j] = 1 - 2.0 * levenshtein(names[i], names[j]) / float(len(names[i]+names[j])) # STEP B: pca dimanesionality reduction to a single-dimension (from the distance space) pca = mlpy.PCA(method='cov') pca.learn(Dnames) coeff = pca.coeff() # STEP C: mapping of 1-dimensional values to colors in a jet-colormap textToColor = pca.transform(Dnames, k=1) textToColor = 255 * (textToColor - textToColor.min()) / (textToColor.max() - textToColor.min()) textmaps = generateColorMap(); colors = [textmaps[int(c)] for c in textToColor] return colors
def RunPCAMlpy(q):
totalTimer = Timer()
# Load input dataset.
Log.Info("Loading dataset", self.verbose)
data = np.genfromtxt(self.dataset, delimiter=',')
try:
with totalTimer:
# Find out what dimension we want.
match = re.search('-d (\d+)', options)
if not match:
k = data.shape[1]
else:
k = int(match.group(1))
if (k > data.shape[1]):
Log.Fatal("New dimensionality (" + str(k) +
") cannot be greater " +
"than existing dimensionality (" +
str(data.shape[1]) + ")!")
q.put(-1)
return -1
# Get the options for running PCA.
s = True if options.find("-s") > -1 else False
# Perform PCA.
prep = mlpy.PCA(whiten=s)
prep.learn(data)
out = prep.transform(data, k)
except Exception as e:
q.put(-1)
return -1
time = totalTimer.ElapsedTime()
q.put(time)
return time
def pcaSvm():
wine = np.loadtxt(r'F:\PY\data\wine.txt', delimiter=',')
x,y = wine[:,1:4],wine[:,0].astype(np.int)
print x.shape, y.shape
pca=mlpy.PCA()
pca.learn(x)
z = pca.transform(x,k=2)
print z.shape
fig1 = plt.figure(1)
title = plt.title('PCA on wine dataset')
plot = plt.scatter(z[:,0],z[:,1],c = y, s = 90, cmap =cm.Reds)
labx = plt.xlabel('First component')
laby = plt.ylabel('Second component')
plt.show()
svm = mlpy.LibSvm(kernel_type='rbf',gamma=20)
svm.learn(z,y)
xmin, xmax = z[:,0].min()-0.1, z[:,0].max()+0.1
ymin, ymax = z[:,1].min()-0.1, z[:,1].max()+0.1
xx, yy = np.meshgrid(np.arange(xmin, xmax, 0.01), np.arange(ymin, ymax, 0.01))
grid = np.c_[xx.ravel(), yy.ravel()]
result = svm.pred(grid)
fig2 = plt.figure(2)
title = plt.title("SVM (rbf kernel) on PCA")
plot1 = plt.pcolormesh(xx, yy, result.reshape(xx.shape), cmap = cm.Greys_r)
plot2 = plt.scatter(z[:, 0], z[:, 1], c=y, s=90, cmap = cm.Reds)
labx = plt.xlabel("First component")
laby = plt.ylabel("Second component")
limx = plt.xlim(xmin, xmax)
limy = plt.ylim(ymin, ymax)
plt.show()
#!/usr/bin/env python # -*- coding: utf-8 -*- #code:[email protected] #7-22.py import numpy as np import matplotlib.pyplot as plt import mlpy np.random.seed(0) mean, cov, n = [0, 0], [[1, 1], [1, 1.5]], 100 x = np.random.multivariate_normal(mean, cov, n) pca = mlpy.PCA() pca.learn(x) coeff = pca.coeff() fig = plt.figure(1) plot1 = plt.plot(x[:, 0], x[:, 1], 'o') plot2 = plt.plot([0, coeff[0, 0]], [0, coeff[1, 0]], linewidth=4, color='r') plot3 = plt.plot([0, coeff[0, 1]], [0, coeff[1, 1]], linewidth=4, color='g') xx = plt.xlim(-4, 4) yy = plt.ylim(-4, 4) z = pca.transform(x, k=1) xnew = pca.transform_inv(z) fig2 = plt.figure(2) plot1 = plt.plot(xnew[:, 0], xnew[:, 1], 'o') xx = plt.xlim(-4, 4) yy = plt.ylim(-4, 4) plt.show()
def visualizeFeaturesFolder(folder, dimReductionMethod, priorKnowledge="none"):
'''
This function generates a chordial visualization for the recordings of the provided path.
ARGUMENTS:
- folder: path of the folder that contains the WAV files to be processed
- dimReductionMethod: method used to reduce the dimension of the initial feature space before computing the similarity.
- priorKnowledge: if this is set equal to "artist"
'''
if dimReductionMethod == "pca":
allMtFeatures, wavFilesList = aF.dirWavFeatureExtraction(
folder, 30.0, 30.0, 0.050, 0.050, computeBEAT=True)
namesCategoryToVisualize = [
ntpath.basename(w).replace('.wav', '').split(" --- ")[0]
for w in wavFilesList
]
namesToVisualize = [
ntpath.basename(w).replace('.wav', '') for w in wavFilesList
]
(F, MEAN, STD) = aT.normalizeFeatures([allMtFeatures])
F = np.concatenate(F)
pca = mlpy.PCA(method='cov') # pca (eigenvalue decomposition)
pca.learn(F)
coeff = pca.coeff()
finalDims = pca.transform(F, k=2)
finalDims2 = pca.transform(F, k=10)
else:
allMtFeatures, Ys, wavFilesList = aF.dirWavFeatureExtractionNoAveraging(
folder, 20.0, 5.0, 0.040, 0.040
) # long-term statistics cannot be applied in this context (LDA needs mid-term features)
namesCategoryToVisualize = [
ntpath.basename(w).replace('.wav', '').split(" --- ")[0]
for w in wavFilesList
]
namesToVisualize = [
ntpath.basename(w).replace('.wav', '') for w in wavFilesList
]
ldaLabels = Ys
if priorKnowledge == "artist":
uNamesCategoryToVisualize = list(set(namesCategoryToVisualize))
YsNew = np.zeros(Ys.shape)
for i, uname in enumerate(
uNamesCategoryToVisualize): # for each unique artist name:
indicesUCategories = [
j for j, x in enumerate(namesCategoryToVisualize)
if x == uname
]
for j in indicesUCategories:
indices = np.nonzero(Ys == j)
YsNew[indices] = i
ldaLabels = YsNew
(F, MEAN, STD) = aT.normalizeFeatures([allMtFeatures])
F = np.array(F[0])
clf = LDA(n_components=10)
clf.fit(F, ldaLabels)
reducedDims = clf.transform(F)
pca = mlpy.PCA(method='cov') # pca (eigenvalue decomposition)
pca.learn(reducedDims)
coeff = pca.coeff()
reducedDims = pca.transform(reducedDims, k=2)
# TODO: CHECK THIS ... SHOULD LDA USED IN SEMI-SUPERVISED ONLY????
uLabels = np.sort(
np.unique((Ys))
) # uLabels must have as many labels as the number of wavFilesList elements
reducedDimsAvg = np.zeros((uLabels.shape[0], reducedDims.shape[1]))
finalDims = np.zeros((uLabels.shape[0], 2))
for i, u in enumerate(uLabels):
indices = [j for j, x in enumerate(Ys) if x == u]
f = reducedDims[indices, :]
finalDims[i, :] = f.mean(axis=0)
finalDims2 = reducedDims
print allMtFeatures.shape
for i in range(finalDims.shape[0]):
plt.text(finalDims[i, 0],
finalDims[i, 1],
ntpath.basename(wavFilesList[i].replace('.wav', '')),
horizontalalignment='center',
verticalalignment='center',
fontsize=10)
plt.plot(finalDims[i, 0], finalDims[i, 1], '*r')
plt.xlim([1.2 * finalDims[:, 0].min(), 1.2 * finalDims[:, 0].max()])
plt.ylim([1.2 * finalDims[:, 1].min(), 1.2 * finalDims[:, 1].max()])
plt.show()
SM = 1.0 - distance.squareform(distance.pdist(finalDims2, 'cosine'))
for i in range(SM.shape[0]):
SM[i, i] = 0.0
chordialDiagram("visualization", SM, 0.50, namesToVisualize,
namesCategoryToVisualize)
SM = 1.0 - distance.squareform(distance.pdist(F, 'cosine'))
for i in range(SM.shape[0]):
SM[i, i] = 0.0
chordialDiagram("visualizationInitial", SM, 0.50, namesToVisualize,
namesCategoryToVisualize)
# plot super-categories (i.e. artistname
uNamesCategoryToVisualize = sort(list(set(namesCategoryToVisualize)))
finalDimsGroup = np.zeros(
(len(uNamesCategoryToVisualize), finalDims2.shape[1]))
for i, uname in enumerate(uNamesCategoryToVisualize):
indices = [
j for j, x in enumerate(namesCategoryToVisualize) if x == uname
]
f = finalDims2[indices, :]
finalDimsGroup[i, :] = f.mean(axis=0)
SMgroup = 1.0 - distance.squareform(
distance.pdist(finalDimsGroup, 'cosine'))
for i in range(SMgroup.shape[0]):
SMgroup[i, i] = 0.0
chordialDiagram("visualizationGroup", SMgroup, 0.50,
uNamesCategoryToVisualize, uNamesCategoryToVisualize)
########################PCA USING SKLEARN#########################
pca = PCA(n_components=2)
pca.fit(vecs_of_nodes)
z = pca.transform(vecs_of_nodes)
print(pca.explained_variance_ratio_)
print(pca.singular_values_)
plt.set_cmap(plt.cm.Paired)
fig1 = plt.figure(1)
title = plt.title("PCA on vecs of nodes")
plot = plt.scatter(z[:, 0], z[:, 1])
labx = plt.xlabel("First component")
laby = plt.ylabel("Second component")
########################PCA USING MLPY#############################
#Dimensionality reduction by Principal Component Analysis (PCA)
pca = mlpy.PCA() # new PCA instance
pca.learn(x) # learn from data
z = pca.transform(x, k=2) # embed x into the k=2 dimensional subspace
z.shape
#Plot the principal components:
plt.set_cmap(plt.cm.Paired)
fig1 = plt.figure(1)
title = plt.title("PCA on iris dataset")
plot = plt.scatter(z[:, 0], z[:, 1], c=y)
labx = plt.xlabel("First component")
laby = plt.ylabel("Second component")
plt.show()
