def giveMeNewTraining(self):
x_tmp = []
y_tmp = []
print("Starting to create new training data ")
for c, files in self.trainingsets.iteritems():
y = self.testsets.keys().index(
c) #We use the testset as a reference
for file in files:
image = cv2.imread(
self.path + c + '/' + file,
cv2.CV_LOAD_IMAGE_GRAYSCALE) #unit8 from e.g. 34 to 255
image2 = expandTraining.distorb(image / 255.)
# cv2.imshow('org', cv2.resize(image, (280, 280)))
# cv2.imshow('mani', cv2.resize(image2, (280, 280)))
# cv2.waitKey(2000)
x_tmp.append(np.reshape(
image2,
len(image2)**2)) #To floats from 0 to 1
y_tmp.append(y)
print("Finished, creating new training data")
perm = np.random.permutation(len(y_tmp))
return theano.shared(np.asarray(x_tmp, theano.config.floatX)[perm],
borrow=True), T.cast(
theano.shared(np.asarray(
y_tmp, theano.config.floatX)[perm],
borrow=True), 'int32')
def giveMeNewTraining():
y_table = np.zeros(10)
y_tmp = []
x_tmp = []
if (show):
cv2.namedWindow('Original', cv2.WINDOW_NORMAL)
cv2.namedWindow('Preprocessed', cv2.WINDOW_NORMAL)
minV = 1e100
maxV = -1e100
c = 0
with gzip.open(filenameValidation) as f:
reader = csv.reader(f)
for row in reader:
y = int(row[0])
y_tmp.append(y)
y_table[y] += 1
vals = np.asarray(row[1:], np.uint8)
n = int(np.sqrt(vals.shape[0]))
img_org = np.reshape(vals/255., (n, n))
import expandTraining
img_dist = expandTraining.distorb(img_org)
img_dist = mask_on_rect2(img_dist)
x_tmp.append(img_dist.reshape(-1))
if show:
cv2.imshow('Original', img_org)
cv2.imshow('Preprocessed', img_dist)
cv2.waitKey(1)
c += 1
print(" Data Range" + str(minV) + " " + str(maxV) + " number of training examples " + str(c))
print(" Balance " + str(y_table))
train_set = np.asarray(x_tmp, theano.config.floatX), np.asarray(y_tmp, theano.config.floatX)
return shared_dataset(train_set, np.random.permutation(train_set[1].shape[0]), testnum=2)
def giveMeNewTraining(self):
x_tmp = []
y_tmp = []
print("Starting to create new training data ")
for c,files in self.trainingsets.iteritems():
y = self.testsets.keys().index(c) #We use the testset as a reference
for file in files:
image = cv2.imread(self.path + c + '/' + file, cv2.CV_LOAD_IMAGE_GRAYSCALE) #unit8 from e.g. 34 to 255
image2 = expandTraining.distorb(image / 255.)
# cv2.imshow('org', cv2.resize(image, (280, 280)))
# cv2.imshow('mani', cv2.resize(image2, (280, 280)))
# cv2.waitKey(2000)
x_tmp.append(np.reshape(image2, len(image2)**2)) #To floats from 0 to 1
y_tmp.append(y)
print("Finished, creating new training data")
perm = np.random.permutation(len(y_tmp))
return theano.shared(np.asarray(x_tmp, theano.config.floatX)[perm],borrow=True), T.cast(theano.shared(np.asarray(y_tmp, theano.config.floatX)[perm],borrow=True), 'int32')
