def test_learning():
crf = IgnoreVoidCRF(n_states=3,
n_features=2,
void_label=2,
inference_method='lp')
ssvm = SubgradientStructuredSVM(crf,
verbose=10,
C=100,
n_jobs=1,
max_iter=50,
learning_rate=0.01)
ssvm.fit(X, Y)
for x in X:
y_hat_exhaustive = exhaustive_inference(crf, x, ssvm.w)
y_hat = crf.inference(x, ssvm.w)
assert_array_equal(y_hat, y_hat_exhaustive)
constr = [
find_constraint(crf, x, y, ssvm.w, y_hat=y_hat)
for x, y, y_hat in zip(X, Y, ssvm.predict(X))
]
losses = [c[3] for c in constr]
slacks = [c[2] for c in constr]
assert_true(np.all(np.array(slacks) >= np.array(losses)))
def test_binary_crf_exhaustive():
# tests qpbo inference against brute force
# on random data / weights
np.random.seed(0)
for i in xrange(10):
x = np.random.uniform(-1, 1, size=(3, 2))
x = np.dstack([-x, np.zeros_like(x)]).copy()
crf = GridCRF(n_features=2, n_states=2)
w = np.random.uniform(-1, 1, size=7)
# check map inference
y_hat = crf.inference(x, w)
y_ex = exhaustive_inference(crf, x, w)
assert_array_equal(y_hat, y_ex)
def test_binary_crf_exhaustive():
# tests qpbo inference against brute force
# on random data / weights
np.random.seed(0)
for i in range(10):
x = np.random.uniform(-1, 1, size=(3, 2))
x = np.dstack([-x, np.zeros_like(x)]).copy()
crf = GridCRF(n_features=2, n_states=2)
w = np.random.uniform(-1, 1, size=7)
# check map inference
y_hat = crf.inference(x, w)
y_ex = exhaustive_inference(crf, x, w)
assert_array_equal(y_hat, y_ex)
def test_inference():
crf = IgnoreVoidCRF(n_states=3, n_features=2, void_label=2,
inference_method='lp')
# set some random weights, do inference and check that everything is ok
np.random.seed(0)
for x, y in zip(X, Y):
w = np.random.normal(size=crf.size_psi)
y_hat, energy = crf.inference(x, w, relaxed=True,
return_energy=True)
energy_svm = np.dot(w, crf.psi(x, y_hat))
assert_almost_equal(energy_svm, -energy)
y_hat_exhaustive = exhaustive_inference(crf, x, w)
y_hat = crf.inference(x, w)
assert_array_equal(y_hat, y_hat_exhaustive)
def test_binary_crf_exhaustive():
# tests graph cut inference against brute force
# on random data / weights
np.random.seed(0)
for i in xrange(50):
x = np.random.uniform(-1, 1, size=(3, 3))
x = np.dstack([-x, np.zeros_like(x)]).copy()
crf = GridCRF()
w = np.random.uniform(-1, 1, size=7)
# check map inference
y_hat = crf.inference(x, w)
y_ex = exhaustive_inference(crf, x, w)
#print(y_hat)
#print(y_ex)
#print("++++++++++++++++++++++")
assert_array_equal(y_hat, y_ex)
def test_inference():
crf = IgnoreVoidCRF(n_states=3,
n_features=2,
void_label=2,
inference_method='lp')
# set some random weights, do inference and check that everything is ok
np.random.seed(0)
for x, y in zip(X, Y):
w = np.random.normal(size=crf.size_psi)
y_hat, energy = crf.inference(x, w, relaxed=True, return_energy=True)
energy_svm = np.dot(w, crf.psi(x, y_hat))
assert_almost_equal(energy_svm, -energy)
y_hat_exhaustive = exhaustive_inference(crf, x, w)
y_hat = crf.inference(x, w)
assert_array_equal(y_hat, y_hat_exhaustive)
def test_learning():
crf = IgnoreVoidCRF(n_states=3, n_features=2, void_label=2,
inference_method='lp')
ssvm = SubgradientStructuredSVM(crf, verbose=10, C=100, n_jobs=1,
max_iter=50, learning_rate=0.01)
ssvm.fit(X, Y)
for x in X:
y_hat_exhaustive = exhaustive_inference(crf, x, ssvm.w)
y_hat = crf.inference(x, ssvm.w)
assert_array_equal(y_hat, y_hat_exhaustive)
constr = [find_constraint(crf, x, y, ssvm.w, y_hat=y_hat)
for x, y, y_hat in zip(X, Y, ssvm.predict(X))]
losses = [c[3] for c in constr]
slacks = [c[2] for c in constr]
assert_true(np.all(np.array(slacks) >= np.array(losses)))
