trainy = np.concatenate((np.zeros(
(1, train_size_boys)), np.ones((1, train_size_girls))),
axis=1)
testy = np.concatenate((np.zeros(
(1, test_size_boys)), np.ones((1, test_size_girls))),
axis=1)
#plt.figure(1)
#show_image.show_image_function(trainx.T, 65, 65)
trainno = np.shape(trainy)[1]
dimno = np.shape(trainx)[0]
testno = np.shape(testy)[1]
plt.figure(1)
plt.ion()
show_image.show_image_function(testx.T, 65, 65)
plt.axis('off')
plt.show()
plt.ioff()
raw_input('press a key to continue ....\n')
plt.close("all")
print 'please wait for Naive Bayes!'
##
# Naive Bayes
# estimate class prior distribution;
py = np.zeros((2, 1))
for i in range(0, 2):
py[i, 0] = np.sum(trainy == i) * 1.0 / trainno
# , data[:, indices_test, 3], data[:, indices_test, 4] \
# , data[:, indices_test, 5], data[:, indices_test, 6] \
# , data[:, indices_test, 7], data[:, indices_test, 8]\
), axis = 1)
testy = np.concatenate((np.zeros((1, test_size)), np.ones((1, test_size)) \
, 2 * np.ones((1, test_size)) \
# , 3 * np.ones((1, test_size)) \
# , 4 * np.ones((1, test_size)), 5 * np.ones((1, test_size)) \
# , 6 * np.ones((1, test_size)), 7 * np.ones((1, test_size)) \
# , 8 * np.ones((1, test_size)), 9 * np.ones((1, test_size)) \
), axis = 1)
testno = np.shape(testy)[1]
plt.figure()
plt.ion()
show_image.show_image_function(testx.T, 16, 16)
plt.axis('off')
plt.show()
plt.ioff()
raw_input('press a key to continue ....\n')
plt.close("all")
print 'please wait for Naive Bayes!'
##
# Naive Bayes
# estimate class prior distribution;
py = np.zeros((num, 1))
for i in range(0,num):
X2 = data[:, :, 2].T
Y = np.concatenate((np.ones((1100, 1)), -np.ones((1100, 1))), axis=0)
H = 16
W = 16
# Create a training set
Xtrain = np.concatenate((X[0:int(1100 * 0.8), :], X[1100:1980, :]), axis=0)
Ytrain = np.concatenate((Y[0:int(1100 * 0.8), :], Y[1100:1980, :]), axis=0)
# test set
Xtest = np.concatenate((X[int(1100 * 0.8):1100, :], X[1980:2200, :]), axis=0)
Ytest = np.concatenate((Y[int(1100 * 0.8):1100, :], Y[1980:2200, :]), axis=0)
plt.figure()
show_image.show_image_function(Xtest, H, W)
plt.axis('off')
plt.title('Test set')
train_size = Ytrain.shape[0]
test_size = Ytest.shape[0]
Xtrain = Xtrain.astype(float)
Xtest = Xtest.astype(float)
print '--running svm\n'
beta, beta0 = svm.svm_function(Xtrain, Ytrain, 1)
Y_hat_train = np.sign(Xtrain.dot(beta.T) + beta0)
precision1 = np.sum((Ytrain == Y_hat_train)
# (2) Both of them have 1100 data points.
# read data
load1 = sio.loadmat('usps_all.mat')
data = load1['data'].astype(float)
# class 1: digit 1
# class 2: digit 2
X = np.concatenate((data[:, :, 0].T, data[:, : ,1].T), axis = 0)
X_row = X.shape[0]
X_col = X.shape[1]
H = 16
W = 16
plt.figure()
show_image.show_image_function(X, H, W)
plt.axis('off')
plt.title('Whole Dataset of Digit 1 and Digit 2')
## Separate the dataset into training and testing
nclass1 = data[:,:,0].shape[1] # 1100
nclass2 = data[:,:,1].shape[1] # 1100
Y = np.concatenate((np.ones((nclass1, 1)), 2 * np.ones((nclass2, 1))), axis = 0)
# Use p percent data as training data
p = 0.8
nclass1_train = np.round(nclass1 * p).astype(int)
nclass1_test = nclass1 - nclass1_train
nclass2_train = np.round(nclass2 * p).astype(int)
nclass2_test = nclass2 - nclass2_train
# Loading the upsc digit dataset
matFile = sio.loadmat('usps_all.mat')
data = matFile['data']
pixelno = data.shape[0]
digitno = data.shape[1]
classno = data.shape[2]
# Displaying the digit 1(data(:,:,1)) and 0(data(:,:,10)) in the dataset
H = 16
W = 16
plt.figure(1)
digits_01 = np.concatenate((np.array(data[:, :, 0]), np.array(data[:, :, 9])),
axis=1).T
show_image.show_image_function(digits_01, H, W)
plt.title('digit 1 and 0')
#plt.figure(1)
#show_image.show_image_function(np.array(data[:,:,9]).T, H, W)
#plt.title('digit 2')
# Extracting the digits 1 and 0 and converting into double
x0 = np.concatenate((np.array(data[:, :, 0]), np.array(data[:, :, 9])), axis=1)
x = np.array(x0, dtype=np.float)
y = np.concatenate((np.ones((1, digitno)), 2 * np.ones((1, digitno))), axis=1)
# number of data points to work with;
m = x.shape[1]
# Normalize the data
Anew = x.T
matFile = sio.loadmat('usps_all.mat')
# data is 256 x 1100 x 10, consisting of
# 1100 16x16 images of 10 digits.
data = matFile['data']
pixelno = data.shape[0]
digitno = data.shape[1]
classno = data.shape[2]
H = 16
W = 16
plt.figure(0)
# Display all images of digits 1 and 0.
digits_01 = np.concatenate((np.array(data[:,:,0]), np.array(data[:,:,9])), axis = 1).T
show_image.show_image_function(digits_01, H, W)
plt.title('digit 1 and 0')
# Create data consisting only 1 and 0.
# x is the images, y is the labels.
x0 = np.array(data[:, :, [0,9]]).reshape((pixelno, digitno * 2))
x = np.array((data[:, :, [0,9]]).reshape((pixelno, digitno * 2)), dtype=np.float)
y = np.concatenate((np.ones((1,digitno)), 2 * np.ones((1,digitno))), axis = 1)
# number of data points to work with;
m = x.shape[1]
###############################################################################
## k-means algorithm;
# Greedy algorithm trying to minimize the objective function;
# A highly vectorized version of kmeans.