scipy.optimize_scalar（）给出“UnboundLocalError:赋值前引用的局部变量'fu”

Question

我正在解决一些消费者的问题，优化一些选择。我正在尝试使用scipy中的optimize.minimize_scalar（）函数，但出现错误：

Traceback (most recent call last):
  File "/Users/User/Documents/GitHub/dynamic_programming/model_our_paper.py", line 238, in <module>
    sol_C, sol_V = model.solve()
  File "/Users/User/Documents/GitHub/dynamic_programming/model_our_paper.py", line 69, in solve
    obj_func, bounds=[0, k], method='bounded'
  File "/Users/User/opt/anaconda3/lib/python3.7/site-packages/scipy/optimize/_minimize.py", line 790, in minimize_scalar
    return _minimize_scalar_bounded(fun, bounds, args, **options)
  File "/Users/User/opt/anaconda3/lib/python3.7/site-packages/scipy/optimize/optimize.py", line 1880, in _minimize_scalar_bounded
    if np.isnan(xf) or np.isnan(fx) or np.isnan(fu):
UnboundLocalError: local variable 'fu' referenced before assignment

我不确定您需要多少代码，因为我想不出一种方法来为我的代码提供一个示例，以便您可以复制。如果您有任何建议，请评论。但引起错误的代码部分如下：

def solve(self):
    shape = (self.T, 2, 50)
    sol_C = np.nan * np.zeros(shape)
    sol_V = np.nan * np.zeros(shape)

    # Last period, consume all.
    for L in (0, 1):
        sol_C[self.T-1, L, :] = self.grid_K.copy()
        sol_V[self.T-1, L, :] = self.utility_f(sol_C[self.T-1, L, :], L)

    # Loop over all other periods.
    for t in range(self.T-2, -1, -1):
        for L in (0, 1):
            for k_i, k in enumerate(self.grid_K):
                obj_func = lambda x: -self.value_of_choice(
                    x, L, k, self.grid_K, sol_V[t+1, :, :]
                )
                result = optimize.minimize_scalar(
                    obj_func, bounds=[0, k], method='bounded'
                )

                sol_C[t, L, k_i] = result.x
                sol_V[t, L, k_i] = result.fun
    return sol_C, sol_V

def value_of_choice(self, x, L, k, K_next, V_next):
    K_plus = k*(1 - self.delta) + L
    EV_next = self.interp_linear_1d_scalar(
        K_next, V_next[L, :], K_plus
    )

    V_guess = self.utility_f(x, L) + self.beta*EV_next
    return V_guess

我尝试了以下方法，效果很好：

from scipy import optimize

f = lambda x: (x - 2)*x*(x - 1)**2
k = 10
res = optimize.minimize_scalar(f, bounds=[0, k], method='bounded')
print(res.x)

我使用的是更新版MacOS的Macbook。尝试更新Scipy，并重新安装了我的python版本，这是我从Anaconda获得的

Answer 1

感谢@rpoleski和@WarrenWeckesser帮助我解决了这个问题。这似乎是边界相等时发生的错误，事实就是如此。这个bug似乎有更新，但它似乎仍在开发中

所以我的工作就是在上限上加一点。有趣的是，这不起作用：

result = optimize.minimize_scalar(
                        obj_func, bounds=[0, k+1.0e-5], method='bounded'
                    )

但是这个起作用了：

result = optimize.minimize_scalar(
                        obj_func, bounds=[0, k+1.0e-4], method='bounded'
                    )