def test_ais_with_semirbm_sanity_check(self):
grbm = GaussianRBM(15, 50)
grbm.b = np.random.randn(grbm.b.shape[0], 1)
grbm.c = np.random.randn(grbm.c.shape[0], 1)
srbm = SemiRBM(50, 20)
srbm.W = srbm.W * 0.
srbm.c = srbm.c * 0.
srbm.L = grbm.W.T * grbm.W
srbm.b = grbm.W.T * grbm.b + grbm.c + 0.5 * np.matrix(np.diag(
srbm.L)).T
srbm.L = srbm.L - np.matrix(np.diag(np.diag(srbm.L)))
ais = Estimator(grbm)
ais.estimate_log_partition_function(num_ais_samples=100,
beta_weights=np.arange(0, 1, 1E-3))
ais = Estimator(srbm)
ais.estimate_log_partition_function(num_ais_samples=100,
beta_weights=np.arange(0, 1, 1E-2))
glogz = grbm._ais_logz + srbm.Y.shape[0] * np.log(2)
slogz = srbm._ais_logz + grbm.X.shape[0] * np.log(np.sqrt(2 * np.pi))
self.assertTrue(np.abs(glogz - slogz) < 1.)
def test_ais_with_gaussianrbm(self): rbm = GaussianRBM(30, 10) rbm.c = np.matrix(np.random.randn(10, 1)) rbm.W = np.matrix(np.random.randn(30, 10)) rbm.b = np.matrix(np.random.rand(30, 1)) ais = Estimator(rbm) ais_logz = ais.estimate_log_partition_function(100, np.arange(0., 1., 1E-4)) brf_logz = utils.logsumexp(rbm._ulogprob_hid(utils.binary_numbers(rbm.Y.shape[0]))) lower = np.log(np.exp(ais_logz) - 4 * np.sqrt(rbm._ais_var)) upper = np.log(np.exp(ais_logz) + 4 * np.sqrt(rbm._ais_var)) self.assertTrue(upper - lower < 1.5) self.assertTrue(lower < brf_logz and brf_logz < upper)
def test_ais_with_semirbm_sanity_check(self): grbm = GaussianRBM(15, 50) grbm.b = np.random.randn(grbm.b.shape[0], 1) grbm.c = np.random.randn(grbm.c.shape[0], 1) srbm = SemiRBM(50, 20) srbm.W = srbm.W * 0. srbm.c = srbm.c * 0. srbm.L = grbm.W.T * grbm.W srbm.b = grbm.W.T * grbm.b + grbm.c + 0.5 * np.matrix(np.diag(srbm.L)).T srbm.L = srbm.L - np.matrix(np.diag(np.diag(srbm.L))) ais = Estimator(grbm) ais.estimate_log_partition_function(num_ais_samples=100, beta_weights=np.arange(0, 1, 1E-3)) ais = Estimator(srbm) ais.estimate_log_partition_function(num_ais_samples=100, beta_weights=np.arange(0, 1, 1E-2)) glogz = grbm._ais_logz + srbm.Y.shape[0] * np.log(2) slogz = srbm._ais_logz + grbm.X.shape[0] * np.log(np.sqrt(2 * np.pi)) self.assertTrue(np.abs(glogz - slogz) < 1.)
def main(argv):
# load preprocessed data samples
print 'loading data...\t',
data = load('../data/vanhateren.npz')
print '[DONE]'
print
# remove DC component (first component)
data_train = data['train'][1:, :]
data_test = data['test'][1:, :]
# create 1st layer
dbn = DBN(GaussianRBM(num_visibles=data_train.shape[0], num_hiddens=100))
# hyperparameters
dbn[0].learning_rate = 5E-3
dbn[0].weight_decay = 1E-2
dbn[0].momentum = 0.9
dbn[0].sigma = 0.65
dbn[0].cd_steps = 1
dbn[0].persistent = True
# train 1st layer
print 'training...\t',
dbn.train(data_train, num_epochs=100, batch_size=100)
print '[DONE]'
# evaluate 1st layer
print 'evaluating...\t',
logptf = dbn.estimate_log_partition_function(num_ais_samples=100,
beta_weights=arange(
0, 1, 1E-3))
loglik = dbn.estimate_log_likelihood(data_test)
print '[DONE]'
print
print 'estimated log-partf.:\t', logptf
print 'estimated log-loss:\t', -loglik / data_test.shape[0] / log(2)
print
def test_ais_with_gaussianrbm(self):
rbm = GaussianRBM(30, 10)
rbm.c = np.matrix(np.random.randn(10, 1))
rbm.W = np.matrix(np.random.randn(30, 10))
rbm.b = np.matrix(np.random.rand(30, 1))
ais = Estimator(rbm)
ais_logz = ais.estimate_log_partition_function(100,
np.arange(0., 1., 1E-4))
brf_logz = utils.logsumexp(
rbm._ulogprob_hid(utils.binary_numbers(rbm.Y.shape[0])))
lower = np.log(np.exp(ais_logz) - 4 * np.sqrt(rbm._ais_var))
upper = np.log(np.exp(ais_logz) + 4 * np.sqrt(rbm._ais_var))
self.assertTrue(upper - lower < 1.5)
self.assertTrue(lower < brf_logz and brf_logz < upper)
def main(argv):
# load preprocessed data samples
print 'loading data...\t',
#data_train, data_test = genData() #load('../data/vanhateren.npz')
data_train, data_test = load_data_BS("data/wavingtree/imagedata_blr_01.pkl")
print '[DONE]'
# remove DC component (first component)
# data_train = data['train'][1:, :]
# data_test = data['test'][1:, :]
# create 1st layer
dbn = DBN(GaussianRBM(num_visibles=data_train.shape[0], num_hiddens=8))
# hyperparameters
dbn[0].learning_rate = 0.001
#dbn[0].weight_decay = 1E-2
dbn[0].momentum = 0.9
dbn[0].sigma = 0.07
dbn[0].cd_steps = 15
# dbn[0].persistent = Trueahsanrafique
# train 1st layer
print 'training...\t',
dbn.train(data_train, num_epochs=20, batch_size=1,shuffle=False)
print '[DONE]'
Ndat = data_test.shape[1]
Nsteps = 1
# evaluate 1st layer
print 'evaluating...\t',
dataout = zeros(120*160)
#datasub = zeros(120*160)
for point in xrange(Ndat):
X = asmatrix(data_test[:,point]).T
dataout = vstack((dataout,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn[0].forward(X) # self.activ(1)
X = dbn[0].backward(Y,X)
#print "S hsape:", X.shape
dataout = vstack((dataout,X.flatten()))
dataout = vstack((dataout,subtract(asarray(X),data_test[:,point],asarray(dbn[0].vsigma))))
# plt.imshow((reshape(data_test[:,point],(120,160))), cmap = cm.Greys_r, interpolation ="nearest")
# plt.axis('off')
# plt.figure(2)
for i in range(10*3):
plt.subplot(10,3,i+1)
plt.imshow((reshape(dataout[i+1,:],(120,160))), cmap = cm.Greys_r, interpolation ="nearest")
plt.axis('off')
plt.figure(2)
for k in range(8):
plt.subplot(4,2,k+1)
plt.imshow((reshape(dbn[0].W[:,k],(120,160))), cmap = cm.Greys_r, interpolation ="nearest")
plt.axis('off')
plt.figure(3)
plt.imshow((reshape(dbn[0].vsigma,(120,160))), cmap = cm.Greys_r, interpolation ="nearest")
plt.show()
# p.figure(1)
# p.plot(data_train[0,:],data_train[1,:], 'b.')
# p.axis([0.0, 1.0, 0.0, 1.0])
# p.figure(2)
# p.plot(data_test[0,:],data_test[1,:], 'b.')
# p.axis([0.0, 1.0, 0.0, 1.0])
# p.figure(3)
# p.plot(datout[0,:],datout[1,:], 'b.')
# p.axis([0.0, 1.0, 0.0, 1.0])
# p.figure(4)
# p.plot(datout[0,:],datout[1,:], 'b.')
#p.axis([0.0, 1.0, 0.0, 1.0])
#p.figure(4)
#p.hist(datout)
print dbn[0].vsigma
p.show()
def main(argv):
# load preprocessed data samples
print 'loading data...\t',
#data_train, data_test = genData() #load('../data/vanhateren.npz')
DIR1 = "data/changedetection\\nightvideos\\tramstation"
data_train_0, data_train_1, data_train_2, data_test = load_data_BS(
DIR1 + "/orignal_color.pkl")
img = Image.open(DIR1 + "/gt001330.png")
x_, y_ = (asarray(img)[:, :]).shape
print "Doie : ", x_, y_
groundtruth = (((asarray(img)[:, :]) / 255.0 > 0.5) * 1).flatten()
#groundtruth = (((asarray(img)[:,:])/255.0 > 0.5) * 1).flatten()
print '[DONE]'
print data_test.shape
# remove DC component (first component)
# data_train = data['train'][1:, :]
# data_test = data['test'][1:, :]
# create 1st layer
dbn = DBN(GaussianRBM(num_visibles=data_train_0.shape[0], num_hiddens=8))
dbn1 = DBN(GaussianRBM(num_visibles=data_train_1.shape[0], num_hiddens=8))
dbn2 = DBN(GaussianRBM(num_visibles=data_train_2.shape[0], num_hiddens=8))
dbn[0].learning_rate = 0.001
dbn1[0].learning_rate = 0.001
dbn2[0].learning_rate = 0.001
# train 1st layer
print 'training...\t',
dbn.train(data_train_0, num_epochs=5, batch_size=1, shuffle=False)
dbn1.train(data_train_1, num_epochs=5, batch_size=1, shuffle=False)
dbn2.train(data_train_2, num_epochs=5, batch_size=1, shuffle=False)
print '[DONE]'
data_test_0 = ((data_test.T)[:, ::3]).T
data_test_1 = ((data_test.T)[:, 1::3]).T
data_test_2 = ((data_test.T)[:, 2::3]).T
# global runtime
# sFilename = "data\\weightlogs"
# f = h5.File(sFileName+".hdf5")
# f.create_group(runtime)
# group.create_dataset('parameters',data=par)
#storeWeights(category,"data\\weightlogs", dbn[0].W, dbn[0].vsigma,dbn1[0].W, dbn1[0].vsigma,dbn2[0].W, dbn2[0].vsigma )
coff = 8
Ndat = 20 #data_test_0.shape[1]
Nsteps = 5
# evaluate 1st layer
print 'evaluating 1...\t',
dataout = zeros(x_ * y_)
# #datasub = zeros(x_*y_)
for point in xrange(Ndat):
#X = asmatrix(data_test_0[:,point]).T
X = asmatrix(data_test_0[:, -1]).T
#dataout = vstack((dataout,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn[0].forward(X) # self.activ(1)
X = dbn[0].backward(Y, X)
#print "S hsape:", X.shape
#dataout = vstack((dataout,X.flatten()))
dataout = vstack((dataout,
subtract(asarray(X), data_test_0[:, -1],
asarray(dbn[0].vsigma), point + 1)))
print 'evaluating 2...\t',
dataout1 = zeros(x_ * y_)
# #datasub = zeros(x_*y_)
for point in xrange(Ndat):
#X = asmatrix(data_test_1[:,point]).T
X = asmatrix(data_test_1[:, -1]).T
#dataout1 = vstack((dataout1,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn1[0].forward(X) # self.activ(1)
X = dbn1[0].backward(Y, X)
#print "S hsape:", X.shape
#dataout1 = vstack((dataout1,X.flatten()))
dataout1 = vstack((dataout1,
subtract(asarray(X), data_test_1[:, -1],
asarray(dbn1[0].vsigma), point + 1)))
print 'evaluating 3...\t',
dataout2 = zeros(x_ * y_)
# #datasub = zeros(x_*y_)
for point in xrange(Ndat):
#X = asmatrix(data_test_2[:,point]).T
X = asmatrix(data_test_2[:, -1]).T
#dataout2 = vstack((dataout2,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn2[0].forward(X) # self.activ(1)
X = dbn2[0].backward(Y, X)
#print "S hsape:", X.shape
#dataout2 = vstack((dataout2,X.flatten()))
dataout2 = vstack((dataout2,
subtract(asarray(X), data_test_2[:, -1],
asarray(dbn2[0].vsigma), point + 1)))
# plt.imshow((reshape(data_test[::3,5],(x_,y_))), cmap = cm.Greys_r, interpolation ="nearest")
# plt.axis('off')
# plt.show()
fmeasureV = zeros(Ndat)
plt.figure(1)
for i in range(Ndat):
plt.subplot(5, 5, i + 1)
d = multiply(asarray(dataout[i + 1, :]), asarray(dataout1[i + 1, :]),
asarray(dataout2[i + 1, :]))
d = mod(d + 1, 2)
open_img = ndimage.median_filter(reshape(d, (x_, y_)), coff)
d = open_img.flatten()
fmeasureV[i] = f_measure(d, groundtruth)
print "Image Example Fmeaure: ", i, " : ", fmeasureV[i] * 100
# d[0::3] = asarray(dataout[i+1,:])
# d[1::3] = asarray(dataout1[i+1,:])
# d[2::3] = asarray(dataout2[i+1,:])
# d[:,:,0] = (reshape(asarray(dataout[i+1,:]),(x_,y_)))
# d[:,:,1] = (reshape(asarray(dataout1[i+1,:]),(x_,y_)))
# d[:,:,2] = (reshape(asarray(dataout2[i+1,:]),(x_,y_)))
# img_s = Image.fromarray(asarray(reshape(d,(x_,y_))*255,dtype="uint8"))
# img_s.save("C:\work\\backgdSubt\dataset\datasets\change detection\\baseline\\baseline\office\grbm" + "\\" + str(i) + ".bmp")
plt.imshow(reshape(d, (x_, y_)),
cmap=cm.Greys_r,
interpolation="nearest")
plt.axis('off')
#plt.figure(2)
plt.show()
DIR2 = "C:\work\\backgdSubt\dataset\datasets\change detection\\nightvideos\\tramstation"
srcDir = DIR2 + "\input"
targetDir = DIR2 + "\grbm"
scalevaraince = input("Enter value: ") #argmax(fmeasureV) + 12
ii = 1
for k in os.listdir(srcDir):
if (ii < 500):
ii = ii + 1
continue
if (ii > 1749):
break
imf = Image.open(srcDir + '/' + k)
print array(imf).shape
data_test = (array(imf)).flatten()
#dataset = np.vstack((dataset,d))
data_test_0 = ((data_test)[::3]).T
data_test_1 = ((data_test)[1::3]).T
data_test_2 = ((data_test)[2::3]).T
Nsteps = 5
# evaluate 1st layer
print 'evaluating ', k, "\t"
# #datasub = zeros(x_*y_)
#for point in xrange(Ndat):
#X = asmatrix(data_test_0[:,point]).T
X = asmatrix(data_test_0).T
#dataout = vstack((dataout,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn[0].forward(X) # self.activ(1)
X = dbn[0].backward(Y, X)
#print "S hsape:", X.shape
#dataout = vstack((dataout,X.flatten()))
dataout = subtract(asarray(X), data_test_0, asarray(dbn[0].vsigma),
scalevaraince + 1)
#X = asmatrix(data_test_1[:,point]).T
X = asmatrix(data_test_1).T
#dataout1 = vstack((dataout1,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn1[0].forward(X) # self.activ(1)
X = dbn1[0].backward(Y, X)
#print "S hsape:", X.shape
#dataout1 = vstack((dataout1,X.flatten()))
dataout1 = subtract(asarray(X), data_test_1, asarray(dbn1[0].vsigma),
scalevaraince + 1)
X = asmatrix(data_test_2).T
#dataout2 = vstack((dataout2,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn2[0].forward(X) # self.activ(1)
X = dbn2[0].backward(Y, X)
#print "S hsape:", X.shape
#dataout2 = vstack((dataout2,X.flatten()))
# plt.imshow(reshape(X,(x_,y_)))
# plt.show()
dataout2 = subtract(asarray(X), data_test_2, asarray(dbn2[0].vsigma),
scalevaraince + 1)
# plt.imshow((reshape(data_test[::3,5],(x_,y_))), cmap = cm.Greys_r, interpolation ="nearest")
# plt.axis('off')
# plt.show()
# plt.figure(1)
# for i in range(Ndat):
# plt.subplot(5,5,i+1)
d = multiply(asarray(dataout), asarray(dataout1), asarray(dataout2))
d = mod(d + 1, 2)
open_img = ndimage.median_filter(
asarray(reshape(d, (x_, y_)) * 255, dtype="uint8"), coff)
#d = open_img.flatten()
#print type(d[0])
# #plt.savefig(targetDir + "\\" + str(ii) + ".png")
#print
img_s = Image.fromarray(open_img)
# plt.figure(2)
# plt.imshow(img_s)#,cmap = cm.Greys_r, interpolation ="nearest")
# plt.show()
if (ii < 1000):
img_s.save(targetDir + "\\bin000" + str(ii) + ".bmp")
else:
img_s.save(targetDir + "\\bin00" + str(ii) + ".bmp")
ii = ii + 1
def test_probabilities(self): grbm = GaussianRBM(7, 13) grbm.W = np.asmatrix(np.random.randn(grbm.X.shape[0], grbm.Y.shape[0])) grbm.b = np.asmatrix(np.random.rand(grbm.X.shape[0], 1)) grbm.c = np.asmatrix(np.random.randn(grbm.Y.shape[0], 1)) grbm.sigma = np.random.rand() * 0.5 + 0.5 grbm.sigma = 1. examples_vis = np.asmatrix(np.random.randn(grbm.X.shape[0], 1000) * 2.) examples_hid = np.matrix(np.random.rand(grbm.Y.shape[0], 100) < 0.5) states_hid = utils.binary_numbers(grbm.Y.shape[0]) # check that conditional probabilities are normalized logprobs = grbm._clogprob_hid_vis(examples_vis, states_hid, all_pairs=True) self.assertTrue(np.all(utils.logsumexp(logprobs, 1) < 1E-8)) # test for consistency logprobs1 = grbm._ulogprob(examples_vis, examples_hid, all_pairs=True) logprobs2 = grbm._clogprob_vis_hid(examples_vis, examples_hid, all_pairs=True) \ + grbm._ulogprob_hid(examples_hid) logprobs3 = grbm._clogprob_hid_vis(examples_vis, examples_hid, all_pairs=True) \ + grbm._ulogprob_vis(examples_vis).T self.assertTrue(np.all(np.abs(logprobs1 - logprobs2) < 1E-10)) self.assertTrue(np.all(np.abs(logprobs1 - logprobs3) < 1E-3))
def test_all_pairs(self): grbm = GaussianRBM(10, 20) grbm.W = np.matrix(np.random.randn(grbm.X.shape[0], grbm.Y.shape[0])) grbm.b = np.matrix(np.random.rand(grbm.X.shape[0], 1)) grbm.c = np.matrix(np.random.randn(grbm.Y.shape[0], 1)) grbm.sigma = np.random.rand() + 0.5 examples_vis = np.matrix(np.random.randn(grbm.X.shape[0], 100)) examples_hid = np.matrix(np.random.rand(grbm.Y.shape[0], 100) < 0.5) logprob1 = grbm._ulogprob(examples_vis, examples_hid) logprob2 = np.diag(grbm._ulogprob(examples_vis, examples_hid, all_pairs=True)) self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-10) logprob1 = grbm._ulogprob(examples_vis[:, 1], examples_hid) logprob2 = grbm._ulogprob(examples_vis, examples_hid, all_pairs=True)[1, :] self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-10) logprob1 = grbm._ulogprob(examples_vis, examples_hid[:, 1]) logprob2 = grbm._ulogprob(examples_vis, examples_hid, all_pairs=True)[:, 1].T self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-10) logprob1 = grbm._clogprob_vis_hid(examples_vis, examples_hid) logprob2 = np.diag(grbm._clogprob_vis_hid(examples_vis, examples_hid, all_pairs=True)) self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-10) logprob1 = grbm._clogprob_vis_hid(examples_vis[:, 1], examples_hid) logprob2 = grbm._clogprob_vis_hid(examples_vis, examples_hid, all_pairs=True)[1, :] self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-10) logprob1 = grbm._clogprob_vis_hid(examples_vis, examples_hid[:, 1]) logprob2 = grbm._clogprob_vis_hid(examples_vis, examples_hid, all_pairs=True)[:, 1].T self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-10) logprob1 = grbm._clogprob_hid_vis(examples_vis, examples_hid) logprob2 = np.diag(grbm._clogprob_hid_vis(examples_vis, examples_hid, all_pairs=True)) self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-5) logprob1 = grbm._clogprob_hid_vis(examples_vis[:, 1], examples_hid) logprob2 = grbm._clogprob_hid_vis(examples_vis, examples_hid, all_pairs=True)[1, :] self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-5) logprob1 = grbm._clogprob_hid_vis(examples_vis, examples_hid[:, 1]) logprob2 = grbm._clogprob_hid_vis(examples_vis, examples_hid, all_pairs=True)[:, 1].T self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-5)
def main(argv):
# load preprocessed data samples
print 'loading data...\t',
#data_train, data_test = genData() #load('../data/vanhateren.npz')
category = "office"
data_train_0,data_train_1,data_train_2, data_test = load_data_BS("..\data\watersurface\orignal_color.pkl")
img = Image.open("..\data\watersurface\hand_segmented_01850.bmp")
x_, y_ = (asarray(img)[:,:,0]).shape
print "Doie : " , (asarray(img)[:,:,0]).shape
groundtruth = (((asarray(img)[:,:,0])/255.0 > 0.5) * 1).flatten()
#groundtruth = (((asarray(img)[:,:])/255.0 > 0.5) * 1).flatten()
print '[DONE]'
print data_test.shape
# remove DC component (first component)
# data_train = data['train'][1:, :]
# data_test = data['test'][1:, :]
# create 1st layer
dbn = DBN(GaussianRBM(num_visibles=data_train_0.shape[0], num_hiddens=8))
dbn1 = DBN(GaussianRBM(num_visibles=data_train_1.shape[0], num_hiddens=8))
dbn2 = DBN(GaussianRBM(num_visibles=data_train_2.shape[0], num_hiddens=8))
dbn[0].learning_rate = 0.001
dbn1[0].learning_rate = 0.001
dbn2[0].learning_rate = 0.001
# train 1st layer
print 'training...\t',
dbn.train(data_train_0, num_epochs=5, batch_size=1,shuffle=False)
dbn1.train(data_train_1, num_epochs=5, batch_size=1,shuffle=False)
dbn2.train(data_train_2, num_epochs=5, batch_size=1,shuffle=False)
print '[DONE]'
data_test_0 = ((data_test.T)[:,::3]).T
data_test_1 = ((data_test.T)[:,1::3]).T
data_test_2 = ((data_test.T)[:,2::3]).T
# global runtime
# sFilename = "data\\weightlogs"
# f = h5.File(sFileName+".hdf5")
# f.create_group(runtime)
# group.create_dataset('parameters',data=par)
#storeWeights(category,"data\\weightlogs", dbn[0].W, dbn[0].vsigma,dbn1[0].W, dbn1[0].vsigma,dbn2[0].W, dbn2[0].vsigma )
Ndat = 25 #data_test_0.shape[1]
Nsteps = 5
# evaluate 1st layer
print 'evaluating 1...\t',
dataout = zeros(x_*y_)
# #datasub = zeros(x_*y_)
for point in xrange(Ndat):
#X = asmatrix(data_test_0[:,point]).T
X = asmatrix(data_test_0[:,-1]).T
#dataout = vstack((dataout,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn[0].forward(X) # self.activ(1)
X = dbn[0].backward(Y,X)
#print "S hsape:", X.shape
#dataout = vstack((dataout,X.flatten()))
dataout = vstack((dataout,subtract(asarray(X),data_test_0[:,-1],asarray(dbn[0].vsigma),point+1)))
print 'evaluating 2...\t',
dataout1 = zeros(x_*y_)
# #datasub = zeros(x_*y_)
for point in xrange(Ndat):
#X = asmatrix(data_test_1[:,point]).T
X = asmatrix(data_test_1[:,-1]).T
#dataout1 = vstack((dataout1,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn1[0].forward(X) # self.activ(1)
X = dbn1[0].backward(Y,X)
#print "S hsape:", X.shape
#dataout1 = vstack((dataout1,X.flatten()))
dataout1 = vstack((dataout1,subtract(asarray(X),data_test_1[:,-1],asarray(dbn1[0].vsigma),point+1)))
print 'evaluating 3...\t',
dataout2 = zeros(x_*y_)
# #datasub = zeros(x_*y_)
for point in xrange(Ndat):
#X = asmatrix(data_test_2[:,point]).T
X = asmatrix(data_test_2[:,-1]).T
#dataout2 = vstack((dataout2,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn2[0].forward(X) # self.activ(1)
X = dbn2[0].backward(Y,X)
#print "S hsape:", X.shape
#dataout2 = vstack((dataout2,X.flatten()))
dataout2 = vstack((dataout2,subtract(asarray(X),data_test_2[:,-1],asarray(dbn2[0].vsigma),point+1)))
# plt.imshow((reshape(data_test[::3,5],(x_,y_))), cmap = cm.Greys_r, interpolation ="nearest")
# plt.axis('off')
# plt.show()
plt.figure(1)
for i in range(Ndat):
plt.subplot(5,5,i+1)
d = multiply(asarray(dataout[i+1,:]),asarray(dataout1[i+1,:]),asarray(dataout2[i+1,:]))
d = mod(d+1,2)
#open_img = ndimage.binary_opening(reshape(d,(x_,y_)))
#close_img = ndimage.binary_closing(open_img)
m_filter = ndimage.median_filter(reshape(d,(x_,y_)),8)
# plt.figure(3)
# plt.imshow(close_img)
# plt.show()
d = m_filter.flatten()
print "Image Example Fmeaure: ",i," : ", f_measure(d,groundtruth) * 100
# d[0::3] = asarray(dataout[i+1,:])
# d[1::3] = asarray(dataout1[i+1,:])
# d[2::3] = asarray(dataout2[i+1,:])
# d[:,:,0] = (reshape(asarray(dataout[i+1,:]),(x_,y_)))
# d[:,:,1] = (reshape(asarray(dataout1[i+1,:]),(x_,y_)))
# d[:,:,2] = (reshape(asarray(dataout2[i+1,:]),(x_,y_)))
#img_s = Image.fromarray(asarray(reshape(d,(x_,y_))*255,dtype="uint8"))
#img_s.save("C:\work\\backgdSubt\dataset\datasets\change detection\\baseline\\baseline\office\grbm" + "\\" + str(i) + ".bmp")
plt.imshow(reshape(d,(x_,y_)), cmap = cm.Greys_r, interpolation ="nearest")
plt.axis('off')
plt.figure(2)
for k in range(8):
plt.subplot(4,2,k+1)
d = zeros((x_*y_*3))
d[0::3] = asarray(dbn[0].W[:,k].flatten())
d[1::3] = asarray(dbn1[0].W[:,k].flatten())
d[2::3] = asarray(dbn2[0].W[:,k].flatten())
plt.imshow(reshape(d,(x_,y_,3)))#, cmap = cm.Greys_r, interpolation ="nearest")
plt.axis('off')
# plt.figure()
# plt.imshow((reshape(dbn[0].vsigma[:19200],(x_,y_))))
# plt.figure(2)
# plt.imshow((reshape(dbn[0].vsigma[19200:19200*2],(x_,y_))))
# plt.figure(3)
# plt.imshow((reshape(dbn[0].vsigma[19200*2:19200*3],(x_,y_))))
plt.figure(3)
print type(dbn[0].vsigma)
plt.imshow(reshape(asarray(dbn[0].vsigma),(x_,y_)))
plt.show()
print dbn[0].vsigma
p.show()
def main(argv):
# load preprocessed data samples
print 'loading data...\t',
#data_train, data_test = genData() #load('../data/vanhateren.npz')
data_train_0, data_train_1, data_train_2, data_test = load_data_BS(
"./data/gf_dborignal_color.pkl")
img = Image.open("./data/gf_db/mensa_seq0_1.1/seq0_0000_0.jpg")
print "Doie : ", (asarray(img)[:, :]).shape
groundtruth = (((asarray(img)[:, :]) / 255.0 > 0.5) * 1).flatten()
#groundtruth = (((asarray(img)[:,:])/255.0 > 0.5) * 1).flatten()
print '[DONE]'
# remove DC component (first component)
# data_train = data['train'][1:, :]
# data_test = data['test'][1:, :]
# create 1st layer
dbn = DBN(GaussianRBM(num_visibles=data_train_0.shape[0], num_hiddens=8))
dbn1 = DBN(GaussianRBM(num_visibles=data_train_1.shape[0], num_hiddens=8))
dbn2 = DBN(GaussianRBM(num_visibles=data_train_2.shape[0], num_hiddens=8))
# train 1st layer
print 'training...\t',
dbn.train(data_train_0, num_epochs=100, batch_size=1, shuffle=False)
dbn1.train(data_train_1, num_epochs=100, batch_size=1, shuffle=False)
dbn2.train(data_train_2, num_epochs=100, batch_size=1, shuffle=False)
print '[DONE]'
data_test_0 = ((data_test.T)[:, ::3]).T
data_test_1 = ((data_test.T)[:, 1::3]).T
data_test_2 = ((data_test.T)[:, 2::3]).T
Ndat = 25 #data_test_0.shape[1]
Nsteps = 5
# evaluate 1st layer
print 'evaluating 1...\t',
dataout = zeros(x_ * y_)
# #datasub = zeros(x_*y_)
for point in xrange(Ndat):
#X = asmatrix(data_test_0[:,point]).T
X = asmatrix(data_test_0[:, -1]).T
#dataout = vstack((dataout,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn[0].forward(X) # self.activ(1)
X = dbn[0].backward(Y, X)
#print "S hsape:", X.shape
#dataout = vstack((dataout,X.flatten()))
dataout = vstack((dataout,
subtract(asarray(X), data_test_0[:, -1],
asarray(dbn[0].vsigma), point + 1)))
print 'evaluating 2...\t',
dataout1 = zeros(x_ * y_)
# #datasub = zeros(x_*y_)
for point in xrange(Ndat):
#X = asmatrix(data_test_1[:,point]).T
X = asmatrix(data_test_1[:, -1]).T
#dataout1 = vstack((dataout1,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn1[0].forward(X) # self.activ(1)
X = dbn1[0].backward(Y, X)
#print "S hsape:", X.shape
#dataout1 = vstack((dataout1,X.flatten()))
dataout1 = vstack((dataout1,
subtract(asarray(X), data_test_1[:, -1],
asarray(dbn1[0].vsigma), point + 1)))
print 'evaluating 3...\t',
dataout2 = zeros(x_ * y_)
# #datasub = zeros(x_*y_)
for point in xrange(Ndat):
#X = asmatrix(data_test_2[:,point]).T
X = asmatrix(data_test_2[:, -1]).T
#dataout2 = vstack((dataout2,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn2[0].forward(X) # self.activ(1)
X = dbn2[0].backward(Y, X)
#print "S hsape:", X.shape
#dataout2 = vstack((dataout2,X.flatten()))
dataout2 = vstack((dataout2,
subtract(asarray(X), data_test_2[:, -1],
asarray(dbn2[0].vsigma), point + 1)))
# plt.imshow((reshape(data_test[::3,5],(x_,y_))), cmap = cm.Greys_r, interpolation ="nearest")
# plt.axis('off')
# plt.show()
plt.figure(1)
for i in range(Ndat):
plt.subplot(5, 5, i + 1)
d = multiply(asarray(dataout[i + 1, :]), asarray(dataout1[i + 1, :]),
asarray(dataout2[i + 1, :]))
d = mod(d + 1, 2)
print "Image Example Fmeaure: ", i, " : ", f_measure(d,
groundtruth) * 100
# d[0::3] = asarray(dataout[i+1,:])
# d[1::3] = asarray(dataout1[i+1,:])
# d[2::3] = asarray(dataout2[i+1,:])
# d[:,:,0] = (reshape(asarray(dataout[i+1,:]),(x_,y_)))
# d[:,:,1] = (reshape(asarray(dataout1[i+1,:]),(x_,y_)))
# d[:,:,2] = (reshape(asarray(dataout2[i+1,:]),(x_,y_)))
plt.imshow(reshape(d, (x_, y_)),
cmap=cm.Greys_r,
interpolation="nearest")
plt.axis('off')
plt.figure(2)
for k in range(8):
plt.subplot(4, 2, k + 1)
d = zeros((x_ * y_ * 3))
d[0::3] = asarray(dbn[0].W[:, k].flatten())
d[1::3] = asarray(dbn1[0].W[:, k].flatten())
d[2::3] = asarray(dbn2[0].W[:, k].flatten())
plt.imshow(reshape(
d, (x_, y_, 3))) #, cmap = cm.Greys_r, interpolation ="nearest")
plt.axis('off')
# plt.figure()
# plt.imshow((reshape(dbn[0].vsigma[:19200],(x_,y_))))
# plt.figure(2)
# plt.imshow((reshape(dbn[0].vsigma[19200:19200*2],(x_,y_))))
# plt.figure(3)
# plt.imshow((reshape(dbn[0].vsigma[19200*2:19200*3],(x_,y_))))
plt.figure(3)
print type(dbn[0].vsigma)
plt.imshow(reshape(asarray(dbn[0].vsigma), (x_, y_)))
plt.show()
print dbn[0].vsigma
p.show()
def main(argv):
# load preprocessed data samples
print 'loading data...\t',
data = load('./data/vanhateren.npz')
print '[DONE]'
print
# remove DC component (first component)
data_train = data['train'][1:, :]
data_test = data['test'][1:, :]
# create 1st layer
dbn = DBN(GaussianRBM(num_visibles=data_train.shape[0], num_hiddens=100))
# hyperparameters
dbn[0].learning_rate = 5E-3
dbn[0].weight_decay = 1E-2
dbn[0].momentum = 0.9
dbn[0].sigma = 0.65
dbn[0].cd_steps = 1
dbn[0].persistent = True
# train 1st layer
print 'training...\t',
dbn.train(data_train, num_epochs=100, batch_size=100)
print '[DONE]'
# evaluate 1st layer
print 'evaluating...\t',
logptf = dbn.estimate_log_partition_function(num_ais_samples=100, beta_weights=arange(0, 1, 1E-3))
loglik = dbn.estimate_log_likelihood(data_test)
print '[DONE]'
print
print 'estimated log-partf.:\t', logptf
print 'estimated log-loss:\t', -loglik / data_test.shape[0] / log(2)
print
# create 2nd layer
dbn.add_layer(SemiRBM(num_visibles=100, num_hiddens=100))
# initialize parameters
dbn[1].L = dbn[0].W.T * dbn[0].W
dbn[1].b = dbn[0].W.T * dbn[0].b + dbn[0].c + 0.5 * asmatrix(diag(dbn[1].L)).T
dbn[1].L = dbn[1].L - asmatrix(diag(diag(dbn[1].L)))
# hyperparameters
dbn[1].learning_rate = 5E-3
dbn[1].learning_rate_lateral = 5E-4
dbn[1].weight_decay = 5E-3
dbn[1].weight_decay_lateral = 5E-3
dbn[1].momentum = 0.9
dbn[1].momentum_lateral = 0.9
dbn[1].num_lateral_updates = 20
dbn[1].damping = 0.2
dbn[1].cd_steps = 1
dbn[1].persistent = True
# train 2nd layer
print 'training...\t',
dbn.train(data_train, num_epochs=100, batch_size=100)
print '[DONE]'
# evaluate 2nd layer
print 'evaluating...\t',
logptf = dbn.estimate_log_partition_function(num_ais_samples=100, beta_weights=arange(0, 1, 1E-3))
loglik = dbn.estimate_log_likelihood(data_test, num_samples=100)
print '[DONE]'
print
print 'estimated log-partf.:\t', logptf
print 'estimated log-loss:\t', -loglik / data_test.shape[0] / log(2)
print
# fine-tune with wake-sleep
dbn[0].learning_rate /= 4.
dbn[1].learning_rate /= 4.
print 'fine-tuning...\t',
dbn.train_wake_sleep(data_train, num_epochs=10, batch_size=10)
print '[DONE]'
# reevaluate
print 'evaluating...\t',
logptf = dbn.estimate_log_partition_function(num_ais_samples=100, beta_weights=arange(0, 1, 1E-3))
loglik = dbn.estimate_log_likelihood(data_test, num_samples=100)
print '[DONE]'
print
print 'estimated log-partf.:\t', logptf
print 'estimated log-loss:\t', -loglik / data_test.shape[0] / log(2)
return 0
def main(argv):
# load preprocessed data samples
print 'loading data...\t',
#data_train, data_test = genData() #load('../data/vanhateren.npz')
#data_train_0,data_train_1,data_train_2, data_test = load_data_BS("data/changedetection\\baseline\highway/orignal_color.pkl")
#groundtruth = (((asarray(img)[:,:])/255.0 > 0.5) * 1).flatten()
print '[DONE]'
# remove DC component (first component)
# data_train = data['train'][1:, :]
# data_test = data['test'][1:, :]
# create 1st layer
dbn = DBN(GaussianRBM(num_visibles=x_ * y_, num_hiddens=8))
dbn1 = DBN(GaussianRBM(num_visibles=x_ * y_, num_hiddens=8))
dbn2 = DBN(GaussianRBM(num_visibles=x_ * y_, num_hiddens=8))
f = h5.File("experiments\weightlogs.hdf5", 'r')
category = "office"
dbn[0].W = (f[category])["weights"][:]
dbn1[0].W = (f[category])["weights1"][:]
dbn2[0].W = (f[category])["weights2"][:]
dbn[0].vsigma = (f[category])["sigma"][:]
dbn1[0].vsigma = (f[category])["sigma1"][:]
dbn2[0].vsigma = (f[category])["sigma2"][:]
# dbn[0].learning_rate = 0.001
# dbn1[0].learning_rate = 0.001
# dbn2[0].learning_rate = 0.001
# train 1st layer
# print 'training...\t',
# dbn.train(data_train_0, num_epochs=1, batch_size=1,shuffle=False)
# dbn1.train(data_train_1, num_epochs=1, batch_size=1,shuffle=False)
# dbn2.train(data_train_2, num_epochs=1, batch_size=1,shuffle=False)
# print '[DONE]'
srcDir = "C:\work\\backgdSubt\dataset\datasets\change detection\\baseline\\baseline\highway\input"
targetDir = "C:\work\\backgdSubt\dataset\datasets\change detection\\baseline\\baseline\highway\grbm"
ii = 1
for k in os.listdir(srcDir):
imf = Image.open(srcDir + '/' + k)
print array(imf).shape
data_test = (array(imf)).flatten()
#dataset = np.vstack((dataset,d))
data_test_0 = ((data_test)[::3]).T
data_test_1 = ((data_test)[1::3]).T
data_test_2 = ((data_test)[2::3]).T
Nsteps = 5
# evaluate 1st layer
print 'evaluating 1...\t',
# #datasub = zeros(x_*y_)
#for point in xrange(Ndat):
#X = asmatrix(data_test_0[:,point]).T
X = asmatrix(data_test_0).T
#dataout = vstack((dataout,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn[0].forward(X) # self.activ(1)
X = dbn[0].backward(Y, X)
#print "S hsape:", X.shape
#dataout = vstack((dataout,X.flatten()))
dataout = subtract(asarray(X), data_test_0, asarray(dbn[0].vsigma), 10)
#X = asmatrix(data_test_1[:,point]).T
X1 = asmatrix(data_test_1).T
#dataout1 = vstack((dataout1,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y1 = dbn1[0].forward(X1) # self.activ(1)
X1 = dbn1[0].backward(Y1, X1)
#print "S hsape:", X.shape
#dataout1 = vstack((dataout1,X.flatten()))
dataout1 = subtract(asarray(X1), data_test_1, asarray(dbn1[0].vsigma),
10)
X2 = asmatrix(data_test_2).T
#dataout2 = vstack((dataout2,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y2 = dbn2[0].forward(X2) # self.activ(1)
X2 = dbn2[0].backward(Y2, X2)
#print "S hsape:", X.shape
#dataout2 = vstack((dataout2,X.flatten()))
# plt.imshow(reshape(X,(x_,y_)))
# plt.show()
dataout2 = subtract(asarray(X2), data_test_2, asarray(dbn2[0].vsigma),
10)
# plt.imshow((reshape(dataout,(x_,y_))), cmap = cm.Greys_r, interpolation ="nearest")
# plt.figure(2)
# plt.imshow((reshape(dataout1,(x_,y_))), cmap = cm.Greys_r, interpolation ="nearest")
# plt.figure(3)
# plt.imshow((reshape(dataout2,(x_,y_))), cmap = cm.Greys_r, interpolation ="nearest")
# plt.axis('off')
# plt.show()
# plt.figure(1)
# for i in range(Ndat):
# plt.subplot(5,5,i+1)
d = multiply(asarray(dataout[:]), asarray(dataout1[:]),
asarray(dataout2[:]))
d = mod(d + 1, 2)
# plt.imshow(reshape(d,(x_,y_)),cmap = cm.Greys_r, interpolation ="nearest")
# print type(d[0])
# #plt.savefig(targetDir + "\\" + str(ii) + ".png")
#print
img_s = Image.fromarray(
asarray(reshape(d, (x_, y_)) * 255, dtype="uint8"))
# plt.figure(2)
# plt.imshow(img_s)#,cmap = cm.Greys_r, interpolation ="nearest")
#plt.show()
img_s.save(targetDir + "\\bin00" + str(ii) + ".bmp")
ii = ii + 1
def test_all_pairs(self):
grbm = GaussianRBM(10, 20)
grbm.W = np.matrix(np.random.randn(grbm.X.shape[0], grbm.Y.shape[0]))
grbm.b = np.matrix(np.random.rand(grbm.X.shape[0], 1))
grbm.c = np.matrix(np.random.randn(grbm.Y.shape[0], 1))
grbm.sigma = np.random.rand() + 0.5
examples_vis = np.matrix(np.random.randn(grbm.X.shape[0], 100))
examples_hid = np.matrix(np.random.rand(grbm.Y.shape[0], 100) < 0.5)
logprob1 = grbm._ulogprob(examples_vis, examples_hid)
logprob2 = np.diag(
grbm._ulogprob(examples_vis, examples_hid, all_pairs=True))
self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-10)
logprob1 = grbm._ulogprob(examples_vis[:, 1], examples_hid)
logprob2 = grbm._ulogprob(examples_vis, examples_hid,
all_pairs=True)[1, :]
self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-10)
logprob1 = grbm._ulogprob(examples_vis, examples_hid[:, 1])
logprob2 = grbm._ulogprob(examples_vis, examples_hid,
all_pairs=True)[:, 1].T
self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-10)
logprob1 = grbm._clogprob_vis_hid(examples_vis, examples_hid)
logprob2 = np.diag(
grbm._clogprob_vis_hid(examples_vis, examples_hid, all_pairs=True))
self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-10)
logprob1 = grbm._clogprob_vis_hid(examples_vis[:, 1], examples_hid)
logprob2 = grbm._clogprob_vis_hid(examples_vis,
examples_hid,
all_pairs=True)[1, :]
self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-10)
logprob1 = grbm._clogprob_vis_hid(examples_vis, examples_hid[:, 1])
logprob2 = grbm._clogprob_vis_hid(examples_vis,
examples_hid,
all_pairs=True)[:, 1].T
self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-10)
logprob1 = grbm._clogprob_hid_vis(examples_vis, examples_hid)
logprob2 = np.diag(
grbm._clogprob_hid_vis(examples_vis, examples_hid, all_pairs=True))
self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-5)
logprob1 = grbm._clogprob_hid_vis(examples_vis[:, 1], examples_hid)
logprob2 = grbm._clogprob_hid_vis(examples_vis,
examples_hid,
all_pairs=True)[1, :]
self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-5)
logprob1 = grbm._clogprob_hid_vis(examples_vis, examples_hid[:, 1])
logprob2 = grbm._clogprob_hid_vis(examples_vis,
examples_hid,
all_pairs=True)[:, 1].T
self.assertTrue(np.abs(logprob1 - logprob2).sum() < 1E-5)
def test_probabilities(self):
grbm = GaussianRBM(7, 13)
grbm.W = np.asmatrix(np.random.randn(grbm.X.shape[0], grbm.Y.shape[0]))
grbm.b = np.asmatrix(np.random.rand(grbm.X.shape[0], 1))
grbm.c = np.asmatrix(np.random.randn(grbm.Y.shape[0], 1))
grbm.sigma = np.random.rand() * 0.5 + 0.5
grbm.sigma = 1.
examples_vis = np.asmatrix(np.random.randn(grbm.X.shape[0], 1000) * 2.)
examples_hid = np.matrix(np.random.rand(grbm.Y.shape[0], 100) < 0.5)
states_hid = utils.binary_numbers(grbm.Y.shape[0])
# check that conditional probabilities are normalized
logprobs = grbm._clogprob_hid_vis(examples_vis,
states_hid,
all_pairs=True)
self.assertTrue(np.all(utils.logsumexp(logprobs, 1) < 1E-8))
# test for consistency
logprobs1 = grbm._ulogprob(examples_vis, examples_hid, all_pairs=True)
logprobs2 = grbm._clogprob_vis_hid(examples_vis, examples_hid, all_pairs=True) \
+ grbm._ulogprob_hid(examples_hid)
logprobs3 = grbm._clogprob_hid_vis(examples_vis, examples_hid, all_pairs=True) \
+ grbm._ulogprob_vis(examples_vis).T
self.assertTrue(np.all(np.abs(logprobs1 - logprobs2) < 1E-10))
self.assertTrue(np.all(np.abs(logprobs1 - logprobs3) < 1E-3))
def main(argv):
# load preprocessed data samples
print 'loading data...\t',
#data_train, data_test = genData() #load('../data/vanhateren.npz')
# data_train_0,data_train_1,data_train_2, data_test = load_data_BS("data/changedetection\intermitentmotion\streetlight/orignal_color.pkl")
img = Image.open("data/changedetection\ptz\continuous/gt000962.png")
print "Doie : ", (asarray(img)[:, :]).shape
groundtruth = (((asarray(img)[:, :]) / 255.0 > 0.5) * 1).flatten()
# #groundtruth = (((asarray(img)[:,:])/255.0 > 0.5) * 1).flatten()
# print '[DONE]'
batches = 3
# remove DC component (first component)
# data_train = data['train'][1:, :]
# data_test = data['test'][1:, :]
# create 1st layer
dbn = DBN(GaussianRBM(num_visibles=x_ * y_, num_hiddens=15))
dbn1 = DBN(GaussianRBM(num_visibles=x_ * y_, num_hiddens=15))
dbn2 = DBN(GaussianRBM(num_visibles=x_ * y_, num_hiddens=15))
dbn[0].learning_rate = 0.0001
dbn1[0].learning_rate = 0.0001
dbn2[0].learning_rate = 0.0001
# train 1st layer
print 'training...\t',
for k in range(2):
print "epoc :", k
for i in range(batches):
global batch_no
batch_no = i
os.path.walk(
'C:\work\\backgdSubt\dataset\changedetection\ptz\continuous\\orignal',
load_batch, 0)
global train_set_x0
global train_set_x1
global train_set_x2
dbn.train(train_set_x0.T,
num_epochs=1,
batch_size=1,
shuffle=False)
dbn1.train(train_set_x1.T,
num_epochs=1,
batch_size=1,
shuffle=False)
dbn2.train(train_set_x2.T,
num_epochs=1,
batch_size=1,
shuffle=False)
print '[DONE]'
os.path.walk(
'C:\work\\backgdSubt\dataset\changedetection\ptz\continuous\\test',
load_test_batch, 0)
data_test_0 = ((data_test.T)[:, ::3]).T
data_test_1 = ((data_test.T)[:, 1::3]).T
data_test_2 = ((data_test.T)[:, 2::3]).T
Ndat = 25 #data_test_0.shape[1]
Nsteps = 5
# evaluate 1st layer
print 'evaluating 1...\t',
dataout = zeros(x_ * y_)
# #datasub = zeros(x_*y_)
for point in xrange(Ndat):
#X = asmatrix(data_test_0[:,point]).T
X = asmatrix(data_test_0[:, -1]).T
#dataout = vstack((dataout,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn[0].forward(X) # self.activ(1)
X = dbn[0].backward(Y, X)
#print "S hsape:", X.shape
#dataout = vstack((dataout,X.flatten()))
dataout = vstack((dataout,
subtract(asarray(X), data_test_0[:, -1],
asarray(dbn[0].vsigma), point + 1)))
print 'evaluating 2...\t',
dataout1 = zeros(x_ * y_)
# #datasub = zeros(x_*y_)
for point in xrange(Ndat):
#X = asmatrix(data_test_1[:,point]).T
X = asmatrix(data_test_1[:, -1]).T
#dataout1 = vstack((dataout1,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn1[0].forward(X) # self.activ(1)
X = dbn1[0].backward(Y, X)
#print "S hsape:", X.shape
#dataout1 = vstack((dataout1,X.flatten()))
dataout1 = vstack((dataout1,
subtract(asarray(X), data_test_1[:, -1],
asarray(dbn1[0].vsigma), point + 1)))
print 'evaluating 3...\t',
dataout2 = zeros(x_ * y_)
# #datasub = zeros(x_*y_)
for point in xrange(Ndat):
#X = asmatrix(data_test_2[:,point]).T
X = asmatrix(data_test_2[:, -1]).T
#dataout2 = vstack((dataout2,X.flatten()))
#print "testing:", X.shape
for recstep in xrange(Nsteps):
Y = dbn2[0].forward(X) # self.activ(1)
X = dbn2[0].backward(Y, X)
#print "S hsape:", X.shape
#dataout2 = vstack((dataout2,X.flatten()))
dataout2 = vstack((dataout2,
subtract(asarray(X), data_test_2[:, -1],
asarray(dbn2[0].vsigma), point + 1)))
# plt.imshow((reshape(data_test[::3,5],(x_,y_))), cmap = cm.Greys_r, interpolation ="nearest")
# plt.axis('off')
# plt.show()
plt.figure(1)
for i in range(Ndat):
plt.subplot(5, 5, i + 1)
d = multiply(asarray(dataout[i + 1, :]), asarray(dataout1[i + 1, :]),
asarray(dataout2[i + 1, :]))
d = mod(d + 1, 2)
#print "Image Example Fmeaure: ",i," : ", f_measure(d,groundtruth) * 100
# d[0::3] = asarray(dataout[i+1,:])
# d[1::3] = asarray(dataout1[i+1,:])
# d[2::3] = asarray(dataout2[i+1,:])
# d[:,:,0] = (reshape(asarray(dataout[i+1,:]),(x_,y_)))
# d[:,:,1] = (reshape(asarray(dataout1[i+1,:]),(x_,y_)))
# d[:,:,2] = (reshape(asarray(dataout2[i+1,:]),(x_,y_)))
plt.imshow(reshape(d, (x_, y_)),
cmap=cm.Greys_r,
interpolation="nearest")
plt.axis('off')
plt.figure(2)
for k in range(15):
plt.subplot(5, 3, k + 1)
d = zeros((x_ * y_ * 3))
d[0::3] = asarray(dbn[0].W[:, k].flatten())
d[1::3] = asarray(dbn1[0].W[:, k].flatten())
d[2::3] = asarray(dbn2[0].W[:, k].flatten())
plt.imshow(reshape(d[0::3], (x_, y_)),
cmap=cm.Greys_r,
interpolation="nearest")
plt.axis('off')
# plt.figure()
# plt.imshow((reshape(dbn[0].vsigma[:19200],(x_,y_))))
# plt.figure(2)
# plt.imshow((reshape(dbn[0].vsigma[19200:19200*2],(x_,y_))))
# plt.figure(3)
# plt.imshow((reshape(dbn[0].vsigma[19200*2:19200*3],(x_,y_))))
plt.figure(3)
print type(dbn[0].vsigma)
plt.imshow(reshape(asarray(dbn[0].vsigma), (x_, y_)))
plt.show()
print dbn[0].vsigma
p.show()
