def test_sigmoid(self):
one_value = npround(base.sigmoid(1))
np_array = npround(base.sigmoid(np.array([2, 3])))
result_array = np.array([0.88080, 0.95257])
multidim_array = npround(base.sigmoid(base.ml2nparray('[-2 0; 4 999999; -1 1]')))
result_array_multidim = base.ml2nparray('[0.11920, 0.50000; 0.98201, 1.00000; 0.26894, 0.73106]')
self.assertEqual(one_value, 0.73106, "base.sigmoid fails with one value")
self.assertTrue((np_array == result_array).all(), "base.sigmoid fails with multiple values in a numpyarray")
self.assertTrue((multidim_array == result_array_multidim).all(), "base.sigmoid fails with multiple values in a multidimensional numpyarray")
def test_sigmoid_gradient(self):
a1 = npround(neuralnetwork.sigmoid_gradient(np.array([2, 3])))
r1 = np.array([0.10499, 0.04518])
a2 = npround(neuralnetwork.sigmoid_gradient([[-2, 0], [4, 999999], [-1, 1]]))
r2 = np.array([[0.10499, 0.25000], [0.01766, 0.0], [0.19661, 0.19661]])
input3 = base.ml2nparray('[-2 0; 4 999999; -1 1]')
a3 = npround(neuralnetwork.sigmoid_gradient(input3))
r3 = base.ml2nparray('[0.10499 0.25000; 0.01766 0.00000; 0.19661 0.19661]')
itr = [(a1, r1), (a2, r2), (a3, r3)]
for answer, sought_result in itr:
self.assertTrue((answer == sought_result).all())
def test_sigmoid_gradient(self):
a1 = npround(neuralnetwork.sigmoid_gradient(np.array([2, 3])))
r1 = np.array([0.10499, 0.04518])
a2 = npround(
neuralnetwork.sigmoid_gradient([[-2, 0], [4, 999999], [-1, 1]]))
r2 = np.array([[0.10499, 0.25000], [0.01766, 0.0], [0.19661, 0.19661]])
input3 = base.ml2nparray('[-2 0; 4 999999; -1 1]')
a3 = npround(neuralnetwork.sigmoid_gradient(input3))
r3 = base.ml2nparray(
'[0.10499 0.25000; 0.01766 0.00000; 0.19661 0.19661]')
itr = [(a1, r1), (a2, r2), (a3, r3)]
for answer, sought_result in itr:
self.assertTrue((answer == sought_result).all())
def test_sigmoid(self):
one_value = npround(base.sigmoid(1))
np_array = npround(base.sigmoid(np.array([2, 3])))
result_array = np.array([0.88080, 0.95257])
multidim_array = npround(
base.sigmoid(base.ml2nparray('[-2 0; 4 999999; -1 1]')))
result_array_multidim = base.ml2nparray(
'[0.11920, 0.50000; 0.98201, 1.00000; 0.26894, 0.73106]')
self.assertEqual(one_value, 0.73106,
"base.sigmoid fails with one value")
self.assertTrue(
(np_array == result_array).all(),
"base.sigmoid fails with multiple values in a numpyarray")
self.assertTrue((multidim_array == result_array_multidim).all(
), "base.sigmoid fails with multiple values in a multidimensional numpyarray"
)
def test_costfunction_no_lambda(self):
theta, X, y = base.ml2nparray('[-1 ; 0.2]', '[1 34 ; 1 35]', '[0 ; 1]')
lmbda = 0
message = "Costfunction without lambda doesn't work properly"
self.cf(theta,
X,
y,
lmbda,
J_result=2.9027,
grad_result=np.array([0.49725, 16.90542]),
message=message)
def test_costfunction_no_lambda2(self):
X = np.array([[-0.2751633, 0.424179 , 0.5403023, -0.1455 , -0.7596879],
[-0.532833 , 0.2836622, 0.7539023, 0.1367372, -0.9576595],
[-0.6536436, 0.9601703, 0.9074468, 0.4080821, -0.9999608],
[-0.8390715, 0.843854 , 0.9887046, -0.5477293, -0.9899925],
[ 0.0044257, 0.6603167, 0.9912028, -0.4161468, -0.9111303]])
theta, y = base.ml2nparray('[1;2;3;4;5]', '[1;0;1;0;1]')
theta = tan(theta)
lmbda = 0
message = "Costfunction without lambda doesn't work properly"
self.cf(theta, X, y, lmbda, J_result=0.6991, grad_result=np.array([-0.09119, -0.01649, 0.05485, -0.01447, -0.07573]), message=message)
def test_costfunction_with_lambda(self):
theta, X, y, = base.ml2nparray('[-1 ; 0.2]', '[1 34 ; 1 35]',
'[0 ; 1]')
lmbda = 1.4
message = "Regularized costfunction with lambda doesn't work properly"
self.cf(theta,
X,
y,
lmbda,
J_result=2.9167,
grad_result=np.array([0.49725, 17.04542]),
message=message)
def test_costfunction_no_lambda2(self):
X = np.array(
[[-0.2751633, 0.424179, 0.5403023, -0.1455, -0.7596879],
[-0.532833, 0.2836622, 0.7539023, 0.1367372, -0.9576595],
[-0.6536436, 0.9601703, 0.9074468, 0.4080821, -0.9999608],
[-0.8390715, 0.843854, 0.9887046, -0.5477293, -0.9899925],
[0.0044257, 0.6603167, 0.9912028, -0.4161468, -0.9111303]])
theta, y = base.ml2nparray('[1;2;3;4;5]', '[1;0;1;0;1]')
theta = tan(theta)
lmbda = 0
message = "Costfunction without lambda doesn't work properly"
self.cf(theta,
X,
y,
lmbda,
J_result=0.6991,
grad_result=np.array(
[-0.09119, -0.01649, 0.05485, -0.01447, -0.07573]),
message=message)
def test_predict(self):
X, theta = base.ml2nparray('[0.3 ; 0.2]', '[1 2.4; 1 -17; 1 0.5]')
testresult = logisticregression.predict(X, theta)
result = np.array([[1], [0], [1]])
self.assertTrue((testresult == result).all(),
'predict function not working properly')
class TestNeuralnetwork(unittest.TestCase):
nnparams = np.array([x / 10 for x in range(1, 19)])
X = cos(base.ml2nparray('[1 2 ; 3 4 ; 5 6]'))
y = base.ml2nparray('[4; 2; 3]')
lmbda = 3
def test_sigmoid_gradient(self):
a1 = npround(neuralnetwork.sigmoid_gradient(np.array([2, 3])))
r1 = np.array([0.10499, 0.04518])
a2 = npround(
neuralnetwork.sigmoid_gradient([[-2, 0], [4, 999999], [-1, 1]]))
r2 = np.array([[0.10499, 0.25000], [0.01766, 0.0], [0.19661, 0.19661]])
input3 = base.ml2nparray('[-2 0; 4 999999; -1 1]')
a3 = npround(neuralnetwork.sigmoid_gradient(input3))
r3 = base.ml2nparray(
'[0.10499 0.25000; 0.01766 0.00000; 0.19661 0.19661]')
itr = [(a1, r1), (a2, r2), (a3, r3)]
for answer, sought_result in itr:
self.assertTrue((answer == sought_result).all())
def test_regularized_costfunction(self):
grad_result = np.array([
0.76614, 0.97990, 0.27246, 0.36416, 0.47507, 0.54614, 0.88342,
0.56876, 0.58467, 0.59814, 1.55931, 1.54462, 1.55631, 1.71189,
1.97834, 1.96892, 1.95977, 2.12233
])
J, grad = neuralnetwork.costfunction(nn_params=self.nnparams,
layer_sizes=[2],
X=self.X,
y=self.y,
lmbda=self.lmbda,
input_layer_size=2,
num_labels=4)
self.assertEqual(int(J * 1000) / 1000, 16.456)
self.assertTrue((npround(grad) == grad_result).all())
J, grad = neuralnetwork.costfunction(nn_params=self.nnparams,
layer_sizes=[2, 2, 4],
X=self.X,
y=self.y,
lmbda=self.lmbda)
self.assertEqual(int(J * 1000) / 1000, 16.456)
self.assertTrue((npround(grad) == grad_result).all())
def test_optimize_and_rand_init(self):
result_theta = np.array([
-7.15389779e-01, -9.18476365e-01, -6.74983060e-06, 4.06540433e-06,
-7.70655209e-06, 4.54637886e-06, -1.24218572e+01, -6.93151748e-01,
-6.93152450e-01, -6.93152026e-01, -1.32334340e-06, 1.08626877e-05,
-9.19086431e-06, -6.83584583e-07, -1.14949724e-06, -2.84416399e-06,
3.05967565e-06, 6.42441311e-07
])
opt_theta = neuralnetwork.nn_optimize([2, 2, 4],
X=self.X,
y=self.y,
lmbda=self.lmbda,
initial_theta=self.nnparams,
method='CG',
epsilon=0.01)
self.assertTrue((npround(result_theta) == npround(opt_theta.x)).all())
def test_costfunction_with_lambda(self):
theta, X, y, = base.ml2nparray('[-1 ; 0.2]', '[1 34 ; 1 35]', '[0 ; 1]')
lmbda = 1.4
message = "Regularized costfunction with lambda doesn't work properly"
self.cf(theta, X, y, lmbda, J_result=2.9167, grad_result=np.array([0.49725, 17.04542]), message=message)
def test_costfunction_no_lambda(self):
theta, X, y = base.ml2nparray('[-1 ; 0.2]', '[1 34 ; 1 35]', '[0 ; 1]')
lmbda = 0
message = "Costfunction without lambda doesn't work properly"
self.cf(theta, X, y, lmbda, J_result=2.9027, grad_result=np.array([0.49725,16.90542]), message=message)
def test_predict(self):
X, theta = base.ml2nparray('[0.3 ; 0.2]', '[1 2.4; 1 -17; 1 0.5]')
testresult = logisticregression.predict(X, theta)
result = np.array([[1], [0], [1]])
self.assertTrue((testresult == result).all(), 'predict function not working properly')