def fit(self, X, Y, T): # X,Y, and T must be sorted by T beforehand
self.t = T
max_iter = 2000
# print(max_iter)
d = X.shape[1]
self.w = np.zeros((d, 1))
f_old = np.inf
for i in range(max_iter):
self.cumulative_h(X, Y)
# h = self.h_estimator()
def nloglf(w):
return NpGlm.nlogl(w, None, self.H, X, Y, T)
self.w, self.f = optimize(nloglf, self.w)
# logging.info('%d\t%f' % (i, self.f / len(T)))
# if self.conv is not None:
# self.conv.append((i, -self.f / len(T)))
if abs(self.f - f_old) < 1e-3:
break
f_old = self.f
def fit(self, X, Y, T):
X = augment(X)
d = X.shape[1]
max_iter = 2000
self.w = np.zeros((d, 1))
f_old = np.inf
for i in range(max_iter):
nloglw = lambda w: WblGlm.nloglw(w, self.a, X, Y, T)
self.w, self.f = optimize(nloglw, self.w)
nlogla = lambda a: WblGlm.nlogla(self.w, a, X, Y, T)
self.a, self.f = optimize(nlogla, self.a)
if abs(self.f - f_old) < 1e-4:
break
f_old = self.f
def fit(self, X, Y, T):
X = augment(X)
d = X.shape[1]
self.w = np.zeros((d, 1))
nloglw = lambda w: WblGlm.nloglw(w, self.a, X, Y, T)
self.w, self.f = optimize(nloglw, self.w)