def test_rt_minimize():
def func(x, theta1, theta2):
f = ((x - theta1 - theta2) ** 2).sum()
df = 2 * (x - theta1 - theta2)
return (f, df)
X = normal(0, 1.0, (3,))
x0 = X[:]
theta1 = asarray([2.3, 1.1, -4.4])
theta2 = asarray([1.0, 2.0, 3.0])
(x, fx, i) = rt_minimize(X, func, -100, theta1, theta2)
assert allclose(x, theta1 + theta2)
assert allclose(fx[-1], 0.0)
def test_rt_minimize():
def func(x, theta1, theta2):
f = ((x - theta1 - theta2)**2).sum()
df = 2 * (x - theta1 - theta2)
return (f, df)
X = normal(0, 1.0, (3, ))
x0 = X[:]
theta1 = asarray([2.3, 1.1, -4.4])
theta2 = asarray([1., 2., 3.])
(x, fx, i) = rt_minimize(X, func, -100, theta1, theta2)
assert allclose(x, theta1 + theta2)
assert allclose(fx[-1], 0.0)
def learn_bocpd(X, model_f, hazard_f, conversion): max_minimize_iter = 30 theta = hazard_f.hazard_params theta.extend(model_f.post_params) theta, nlml, _ = rt_minimize(theta, bocpd_dwrap1D, -max_minimize_iter, X, model_f, hazard_f, conversion, hazard_f.num_hazard_params) print "we make it to here!" hazard_f.hazard_params = theta[0:num_hazard_params] model_f.model_params = theta[num_hazard_params:] hazard_f.hazard_params[0] = logistic(hazard_f.hazard_params[0]) model_f.model_params[1:] = np.exp(model_f.model_params[1:]) return hazard_params, model_params, nlml
def learn_bocpd(X, model_f, hazard_f, conversion):
max_minimize_iter = 30
theta = hazard_f.hazard_params
theta.extend(model_f.post_params)
theta, nlml, _ = rt_minimize(theta, bocpd_dwrap1D, -max_minimize_iter, X,
model_f, hazard_f, conversion,
hazard_f.num_hazard_params)
print "we make it to here!"
hazard_f.hazard_params = theta[0:num_hazard_params]
model_f.model_params = theta[num_hazard_params:]
hazard_f.hazard_params[0] = logistic(hazard_f.hazard_params[0])
model_f.model_params[1:] = np.exp(model_f.model_params[1:])
return hazard_params, model_params, nlml
