def main():
svm_params = dict(svm_type=cv2.SVM_C_SVC,
kernel_type=cv2.SVM_LINEAR,
degree=None,
gamma=5.383,
C=1)
# Creates visual platrorm
vis = Visual()
# User input ################################################################
ans = '1' # raw_input("Would like to create model (1) or to view outputs (2) : ")
if ans == '1':
print "Reading data set : "
raw_img, raw_lbl = MyMnist.read()
print "Done \n"
lbl_img = zip(raw_lbl, raw_img)
lbl_img = shuffle(lbl_img)
raw_lbl = np.asarray([t[0] for t in lbl_img])
raw_img = np.asarray([t[1] for t in lbl_img]).reshape(
(raw_lbl.size, config.height * config.width))
print "Now lets start training : \n"
# User input ################################################################
is_hog = 'y' # raw_input("Would you like to use Hog (y/n) : ")
if (is_hog == 'y'):
hog_scale_w = int(raw_input("What width would you like to use : "))
hog_scale_h = int(
raw_input("What height would you like to use : "))
hog_degrees = int(
raw_input("What amount of degrees would you like to use : "))
hog_data = np.zeros([
raw_lbl.size, config.height * config.width / hog_scale_h /
hog_scale_w * hog_degrees
], 'float32')
for i in range(0, raw_lbl.size):
hog_data[i] = Hog.getimghog(
raw_img[i].reshape((config.height, config.width)),
[[1, 0, -1], [2, 0, -1], [1, 0, -1]],
[[1, 2, 1], [0, 0, 0], [-1, -2, -1]], hog_scale_h,
hog_scale_w, hog_degrees)
raw_img = hog_data
# User input ################################################################
out = 2 # int(raw_input("Would you like to use Linear (1) or Kernel (2) :"))
svm_params['C'] = float(raw_input("Please enter C parameter : "))
if out == 1:
svm_params['kernel_type'] = cv2.SVM_LINEAR
else:
svm_params['kernel_type'] = cv2.SVM_POLY
svm_params['degree'] = int(
2) # float(raw_input("Please enter SVM degree parameter : "))
svm_params['gamma'] = float(raw_input("Please enter gamma : "))
name = "HOG " + str(hog_scale_h) + "x" + str(hog_scale_w) + " d" + str(
hog_degrees) + " C" + str(svm_params['C']) + " G" + str(
svm_params['gamma']
) #raw_input("How would you like to name your model : ")
step = int(round(raw_lbl.size * 0.1)) - 1
test_lbls = list()
pred_lbls = list()
for i in range(0, 10):
print "Start " + str(i) + " training ",
train_img = np.concatenate((raw_img[:i, :], raw_img[i + step:, :]))
train_lbl = np.append(raw_lbl[:i], raw_lbl[i + step:])
test_img = raw_img[i:i + step, :]
test_lbl = raw_lbl[i:i + step]
test_lbls.append(test_lbl)
svm = cv2.SVM()
svm.train(train_img, train_lbl, params=svm_params)
print "=> Predicting",
pred_lbls.append(svm.predict_all(test_img))
print "=> Done"
s = SvmOutput(name, svm_params, svm,
np.array(test_lbls).ravel(),
np.array(pred_lbls).ravel())
s.save()
s.showdata()
else:
ans = raw_input(
"Would you like to view a specific model (1) \nor the roc curve of a few models (2) :"
)
models_dir = os.walk(config.default_path).next()[1]
if ans == '1':
print "Available models : "
for i in range(1, len(models_dir) + 1):
print "\t(" + str(i) + ") " + models_dir[i - 1]
ans = int(raw_input("Which model would you like to view : "))
model = SvmOutput(name=models_dir[ans - 1], readfile=True)
model.showdata()
else:
roc_models = list()
ans = 'y'
while (ans == 'y') & (len(models_dir) > 0):
print "Available models : "
for i in range(1, len(models_dir) + 1):
print "\t(" + str(i) + ") " + models_dir[i - 1]
model_add = int(
raw_input("Which model would you like to view : "))
roc_models.append(
SvmOutput(name=models_dir[model_add - 1], readfile=True))
ans = raw_input(
"Would you like to choose another model (y/n) : ")
models_dir.pop(model_add - 1)
vis.showROC(roc_models)
