我正在python中建立一个oneVsAll分类器，并使用scipy.optimize.fmin_cg获得theta向量的最佳值。这是我的分类器功能。
def oneVsAll(X, y, K, reg_parameter):
X = np.hstack((np.ones((X.shape[0],1)), X))
theta = initialiseTheta((K,X.shape[1]))
print(theta[0].shape)
for i in range(K):
print("In for loop")
digit_class = i if i else 10
theta[i] = opt.fmin_cg(f = lrCostFunction, fprime=None, x0=theta[i], args = (X, (y==digit_class).flatten() , reg_parameter), maxiter=50)
return theta
这是我的成本函数
def lrCostFunction(theta, X, y, reg_parameter):
m = y.shape[0] #number of training examples
J =0
grad = np.zeros(theta.shape)
Z = np.dot(X, theta)
print("Z shape", Z.shape)
hx = sigmoid(Z)
print("hx - y shape", (hx-y).shape)
print("X shape",(X[:,0].T).shape )
print("dot product shape", np.dot(X[:,0].T, (hx -y)))
J = -(1/m)*(np.sum(np.multiply(y, np.log(hx)) + np.multiply((1-y), np.log(1-hx)))) + (reg_parameter/(2*m))*(np.sum(np.power(theta[1:], 2)))
grad_unregularized = (1/m)*(np.dot(X[:,1:].T, (hx -y)))
grad[0] = (1/m)*(np.dot(X[:,0].T, (hx -y)))
grad[1:] = grad_unregularized + (reg_parameter/m)*(theta[1:])
return (J, grad)