def __init__(self, bias=False, learn_rate=None, convergence=None, regularization=None):
self.ws = np.array([])
self.bias = bias
self.learn_rate = learn_rate
self.convergence = convergence
self.l = regularization
self.hfunc = lambda x, y: sigmoid(x @ y)
self.costfunc = log_likelihood
def optimize(self, xs, ys):
m, n = xs.shape
ws = np.zeros(len(xs[0]))
iters = 0
grad = 1
while iters < self.max_iters: #and np.linalg.norm(sigmoid(xs @ ws) - ys) > self.convergence:
grad = (expit(xs @ ws) - ys) @ xs + self.r * ws
ws -= self.alpha * grad
iters += 1
print('Gradient descent finished: {} iters, cost = {}'.format(iters, \
np.linalg.norm(sigmoid(xs @ ws) - ys)))
return ws
