def LDA_B(rootdir, folderA, folderB, folderC,ft):
pos = []
neg = []
imgspos = []
imgsneg = []
if 0 == cmp(ft, 'gabor'):
print 'feature type: GABOR'
gbf = igbf.GABOR_FEAT()
elif 0 == cmp(ft, 'hog'):
print 'feature type: HOG'
gbf = ihogf.HOG_FEAT()
elif 0 == cmp(ft, 'lbp'):
print 'feature type: LBP'
gbf = ilbpf.LBP_FEAT()
elif 0 == cmp(ft, 'dwt'):
print 'feature type: DWT'
gbf = idwtf.DWT_FEAT()
elif 0 == cmp(ft, 'yuv'):
print 'feature type: YUV'
gbf = iyuvf.YUV_FEAT()
else:
print 'unknown feature type'
return
#1--class A
fvs, imgs = gbf.gen_folder(os.path.join(rootdir, folderA), 1000)
pos.extend(fvs)
imgspos.extend(imgs)
#2--class B
fvs, imgs = gbf.gen_folder(os.path.join(rootdir, folderB), 1000)
neg.extend(fvs)
imgsneg.extend(imgs)
#3--train
label0 = [0 for k in range(len(pos))]
label1 = [1 for k in range(len(neg))]
samples = np.array(pos + neg)
labels = np.array(label0 + label1)
imgs = imgspos + imgsneg
print 'before pca : ', samples.shape
# com_numx = 300
# if com_numx + 10 > len(imgs):
# com_numx = len(imgs) - 10
clf_pca = PCA(0.95)
samples = clf_pca.fit_transform(samples)
print 'after pca : ', samples.shape
clf_lda = LDA()
clf_lda.fit(samples,labels)
#4--predict
fvs, imgs = gbf.gen_folder(os.path.join(rootdir, folderC), 100000)
samples = np.array(fvs)
samples = clf_pca.transform(samples)
cnf = clf_lda.decision_function(samples)
X = []
for k in range(len(imgs)):
X.append((cnf[k], imgs[k]))
X = sorted(X, key = lambda a : a[0])
lineA = "" #sometimes, the positive set is split into two parts
lineB = ""
for cnf, img in X:
if cnf > 0:
lineA += img + '\n'
else:
lineB += img + '\n'
with open('A.txt', 'w') as f:
f.writelines(lineA)
with open('B.txt', 'w') as f:
f.writelines(lineB)
return
XX=[]
for i in xrange(len(Y)):
if Y[i]==value:
XX.append(X[i])
return XX
out=open(sys.argv[1],"r")
model=LDA()
X, Y = read_fea(sys.argv[1])
sel = VarianceThreshold(threshold=0)
model.fit(sel.fit_transform(X), Y)
warning("useful features dim: "+str(len(sel.get_support(True))))
if hasattr(model,'score'):
warning("accuracy on training set: "+str(model.score(sel.transform(X), Y)))
if len(sys.argv)>2:
X, Y = read_fea(sys.argv[2])
warning("accuracy on cv set: "+str(model.score(sel.transform(X), Y)))
if len(sys.argv)>3:
X, Y = read_fea(sys.argv[3])
warning("accuracy on dev set: "+str(model.score(sel.transform(X), Y)))
if len(sys.argv)>4:
ref = model.decision_function(sel.transform(X))
X, Y = read_fea(sys.argv[4], True)
Z = model.decision_function(sel.transform(X))
Z = (Z-ref.mean(axis=0)[np.newaxis,:])/ref.std(axis=0)[np.newaxis,:]
for i in xrange(len(Y)):
ZZ=np.array(Z[i][1:])
print('S'+str(Y[i])+' '+str(Z[i][0]))
def LDA_B(rootdir, folderA, folderB, folderC, ft):
pos = []
neg = []
imgspos = []
imgsneg = []
if 0 == cmp(ft, 'gabor'):
print 'feature type: GABOR'
gbf = igbf.GABOR_FEAT()
elif 0 == cmp(ft, 'hog'):
print 'feature type: HOG'
gbf = ihogf.HOG_FEAT()
elif 0 == cmp(ft, 'lbp'):
print 'feature type: LBP'
gbf = ilbpf.LBP_FEAT()
elif 0 == cmp(ft, 'dwt'):
print 'feature type: DWT'
gbf = idwtf.DWT_FEAT()
elif 0 == cmp(ft, 'yuv'):
print 'feature type: YUV'
gbf = iyuvf.YUV_FEAT()
else:
print 'unknown feature type'
return
#1--class A
fvs, imgs = gbf.gen_folder(os.path.join(rootdir, folderA), 1000)
pos.extend(fvs)
imgspos.extend(imgs)
#2--class B
fvs, imgs = gbf.gen_folder(os.path.join(rootdir, folderB), 1000)
neg.extend(fvs)
imgsneg.extend(imgs)
#3--train
label0 = [0 for k in range(len(pos))]
label1 = [1 for k in range(len(neg))]
samples = np.array(pos + neg)
labels = np.array(label0 + label1)
imgs = imgspos + imgsneg
print 'before pca : ', samples.shape
# com_numx = 300
# if com_numx + 10 > len(imgs):
# com_numx = len(imgs) - 10
clf_pca = PCA(0.95)
samples = clf_pca.fit_transform(samples)
print 'after pca : ', samples.shape
clf_lda = LDA()
clf_lda.fit(samples, labels)
#4--predict
fvs, imgs = gbf.gen_folder(os.path.join(rootdir, folderC), 100000)
samples = np.array(fvs)
samples = clf_pca.transform(samples)
cnf = clf_lda.decision_function(samples)
X = []
for k in range(len(imgs)):
X.append((cnf[k], imgs[k]))
X = sorted(X, key=lambda a: a[0])
lineA = "" #sometimes, the positive set is split into two parts
lineB = ""
for cnf, img in X:
if cnf > 0:
lineA += img + '\n'
else:
lineB += img + '\n'
with open('A.txt', 'w') as f:
f.writelines(lineA)
with open('B.txt', 'w') as f:
f.writelines(lineB)
return
XX.append(X[i])
return XX
out = open(sys.argv[1], "r")
model = LDA(solver='lsqr')
X, Y = read_fea(sys.argv[1])
sel = VarianceThreshold(threshold=0)
model.fit(sel.fit_transform(X), Y)
warning("useful features dim: " + str(len(sel.get_support(True))))
if hasattr(model, 'score'):
warning("accuracy on training set: " +
str(model.score(sel.transform(X), Y)))
if len(sys.argv) > 2:
X, Y = read_fea(sys.argv[2])
warning("accuracy on cv set: " + str(model.score(sel.transform(X), Y)))
if len(sys.argv) > 3:
X, Y = read_fea(sys.argv[3])
warning("accuracy on dev set: " +
str(model.score(sel.transform(X), Y)))
if len(sys.argv) > 4:
ref = model.decision_function(sel.transform(X))
X, Y = read_fea(sys.argv[4], True)
Z = model.decision_function(sel.transform(X))
Z = (Z - ref.mean(axis=0)[np.newaxis, :]) / ref.std(axis=0)[np.newaxis, :]
for i in xrange(len(Y)):
ZZ = np.array(Z[i][1:])
print('S' + str(Y[i]) + ' ' + str(Z[i][0]))
