def get_flip_error(val):
while True:
pos = random.randint(0, 19)
error = bit_flip(val, pos)
if not math.isnan(error) and not math.isinf(error):
break
return error
def preprocess_for_classifier(dataset, N=10):
dataset = np.vstack([dataset] * N) # copy the dataset N times
has_error = np.zeros(len(dataset))
sign = 1
THREASHOLD = 0.5
for i in range(len(dataset)):
if random.randint(0, 1):
x = random.randint(0, dataset[i].shape[0] - 1)
y = random.randint(0, dataset[i].shape[1] - 1)
has_error[i] = 1
'''
d = dataset[i][x,y]
error = random.uniform(0.01*d, 0.9*d)
sign = 1 if random.randint(0,1) == 1 else -1
dataset[i][x,y] = error
'''
bit_pos = random.randint(1, 10)
#bit_pos = 8
error = bit_flip(dataset[i][x, y], bit_pos)
if math.isnan(error) or math.isinf(error):
has_error[i] = 0
continue
#if error < 0.0001:
# error = 0.0001
#if error > 5*d:
# error = 5*d
#print 'old:', dataset[i][x,y], ', pos:', bit_pos, ', new:', error
dataset[i][x, y] = error
return dataset, has_error
def get_flip_error(val):
while True:
pos = random.randint(0, 20)
error = bit_flip(val, pos)
if not math.isnan(error) and not math.isinf(error):
break
error = min(10e+5, error)
error = max(-10e+5, error)
return error
def get_flip_error(val):
tmp = val
# flip bits 30 times
for i in range(30):
while True:
pos = random.randint(0, 63)
tmp = bit_flip(tmp, pos)
if not math.isnan(tmp) and not math.isinf(tmp):
break
error = tmp
return error