def _test_dsk(self, input_shape):
input_array = np.random.uniform(size=input_shape)
print("[+] Starting dsk computation...")
with timeit() as timestats:
out = self.compute_features(input_array)
print("[+] Dsk computation done...")
print("Input shape: {} || Output shape: {}".format(
input_shape, out.shape))
print("Elapsed time: {}".format(timestats.elapsed_time))
def _test_presmoothing(self, input_shape):
input_array = np.random.uniform(size=input_shape)
with timeit() as timestats:
presmoothed = self.presmoothing(input_array)
print("Input shape: {} || Output shape: {}".format(
input_shape, presmoothed.size()))
print("Elapsed time on {}: {}".format(self.device,
timestats.elapsed_time))
def _test_gradient(self, input_shape, wrt='0'):
input_array = np.random.uniform(size=input_shape)
# Presmooth
presmoothed = self.presmoothing(input_array)
# Get gradient func to test
grad_func = getattr(self, "d{}".format(wrt))
# Time gradient
with timeit() as timestats:
g = grad_func(presmoothed)
print("Input shape: {} || Output shape: {}".format(
presmoothed.size(), g.size()))
print("Elapsed time: {}".format(timestats.elapsed_time))
def _test_dmag_2d(self, input_shape):
input_array = np.random.uniform(size=input_shape)
# Presmooth
presmoothed = self.presmoothing(input_array)
# Compute gradients
g0 = self.d0(presmoothed)
g1 = self.d1(presmoothed)
# Compute gradient magnitude
with timeit() as timestats:
gmag = self.dmag(g0, g1)
print("Input shape: {} || Output shape: {}".format(
presmoothed.size(), gmag.size()))
print("Elapsed time: {}".format(timestats.elapsed_time))
def _test_laplacian_2d(self, input_shape):
input_array = np.random.uniform(size=input_shape)
# Presmooth
presmoothed = self.presmoothing(input_array)
# Compute gradients
g0 = self.d0(presmoothed)
g1 = self.d1(presmoothed)
g00 = self.d0(g0)
g11 = self.d1(g1)
# Compute and time laplacian
with timeit() as timestats:
lap = self.laplacian(g00, g11)
print("Input shape: {} || Output shape: {}".format(
presmoothed.size(), lap.size()))
print("Elapsed time: {}".format(timestats.elapsed_time))
def conv(input_size, kernel_size, num_output=1, num_input=1):
input_variable = Variable(torch.from_numpy(np.random.uniform(size=(num_input, 1,
input_size[0],
input_size[1]))))
kernel_variable = Variable(torch.from_numpy(np.random.uniform(size=(num_output,
1,
kernel_size[0],
kernel_size[1]))))
# bias = Variable(torch.from_numpy(np.zeros(shape=(num_output,))))
with timeit() as timestat:
# output = conv2d(input_variable, kernel_variable, bias=bias, groups=num_output)
output = conv2d(input_variable, kernel_variable)
print("Input size: {} || Kernel size: {} || Num outputs: {}".format(input_size, kernel_size,
num_output))
print("Output shape: {}".format(output.data.size()))
print("Elapsed time: {}".format(timestat.elapsed_time))
def _test_eighess_3d(self, input_shape):
input_array = np.random.uniform(size=input_shape)
# Presmooth
presmoothed = self.presmoothing(input_array)
# Compute gradients
g0 = self.d0(presmoothed)
g1 = self.d1(presmoothed)
g2 = self.d2(presmoothed)
g00 = self.d0(g0)
g01 = self.d1(g0)
g02 = self.d2(g0)
g11 = self.d1(g1)
g12 = self.d2(g1)
g22 = self.d2(g2)
# Compute and time laplacian
with timeit() as timestats:
eighess = self.eighess(g00, g01, g02, g11, g12, g22)
print("Input shape: {} || Output shape: {}".format(
presmoothed.size(), eighess.size()))
print("Elapsed time: {}".format(timestats.elapsed_time))
# Create a random dataset
rng = np.random.RandomState(1)
X = np.sort(200 * rng.rand(600, 1) - 100, axis=0)
y = np.array([
0
if x > X.mean() or np.random.choice([True, False, False, False]) else 1
for x in X
]).T
X_train, X_test, y_train, y_test = train_test_split(X,
y,
train_size=400,
random_state=4)
dsk = {
"classifier_type": "RandomForest",
"RF": (learning.get_classifier, "classifier_type"),
"X_train": X_train,
"X_test": X_test,
"y_train": y_train,
"y_test": y_test,
'trained-RF': (learning.train, 'RF', "X_train", "y_train"),
'predict': (learning.predict, 'trained-RF', 'X_test')
}
with timeit() as time_info:
output = get(dsk, 'predict')
print("Time Elapsed: {}".format(time_info.elapsed_time))