def _gradient_descent(self, X, y):
X = add_intercept(X, 1)
self.w = np.ones((1, self.n_features + 1))
for it in range(self._max_iters):
grad = self._cost(X, y, self.w)
self.w -= (self._eta + 1 / (2 + it)) * grad
# print("iter =", it, "w =", self.w, "grad =", grad)
self.w = self.w.flatten()
def _stochastic_gradient_descent(self, X, y):
step = 0
self.n_features = X.shape[1]
self.n_samples = X.shape[0]
self.w = np.zeros((1, self.n_features))
self.w = add_intercept(self.w, 0)
X = add_intercept(X, 1)
while not self._max_iters or step < self._max_iters:
step += 1
update_pos = self.classify(X, y)
if update_pos >= 0:
self.update(X, y, update_pos)
print("Itetator {0}, Update Pos: {1}, w: {2}".format(
step, update_pos, self.w))
else:
break
return self.w
def predict(self, X):
return np.sign(np.dot(add_intercept(X, 1), self.w.T)).flatten()
def predict(self, X):
X = add_intercept(X, 1)
s = np.dot(self.w, X.T)
return self.sigmod(s)