Question

我正在尝试编写一个可以被numba矢量化的函数，作为一组数组的ufunc：

from numba import float32, float64, vectorize

@vectorize([float32, float32, float32, float32(float32, float32, float32, float32, float32, float32, float32, numba.float32),
            float64, float64, float64, float64(float64, float64, float64, float64, float64, float64, float64, numba.float64)])
def _cafec_coeff_ufunc(ET_bar,
                       PET_bar,
                       R_bar,
                       PR_bar,
                       L_bar,
                       PL_bar,
                       RO_bar,
                       PRO_bar):

    # calculate alpha
    if PET_bar == 0:
        if ET_bar == 0:
            alpha = 1
        else:
            alpha = 0
    else:
        alpha = ET_bar / PET_bar

    # calculate beta
    if PR_bar == 0:
        if R_bar == 0:
            beta = 1
        else:
            beta = 0
    else:
        beta = R_bar / PR_bar

    # calculate gamma
    if PRO_bar == 0:
        if RO_bar == 0:
            gamma = 1
        else:
            gamma = 0
    else:
        gamma = RO_bar / PRO_bar

    # calculate delta
    if PL_bar == 0:
        if L_bar == 0:
            delta = 1
        else:
            delta = 0
    else:
        delta = L_bar / PL_bar

    return alpha, beta, gamma, delta

我在另一个函数中调用上面的函数，传递它需要的八个数组作为参数。我的期望是这个函数将以广播方式应用，就好像每个元素的1-D循环i（假设所有输入数组大小相同），ufunc从第i个索引的每个数组得到标量作为其参数，最后返回四个数组，在输出数组的第i个位置具有相应的值（ufunc广播的结果缩减为输出数组）。

alpha, beta, gamma, delta = _cafec_coeff_ufunc(ETbar, PETbar, Rbar, PRbar, Lbar, PLbar, RObar, PRObar)

当我将上面的代码作为调用方法的单元测试的一部分运行时，我在控制台中遇到以下错误：

  File "C:\home\git\indices_python\palmer.py", line 342, in <module>
    numba.float64, numba.float64, numba.float64, numba.float64(numba.float64, numba.float64, numba.float64, numba.float64, numba.float64, numba.float64, numba.float64, numba.float64)])
  File "C:\Anaconda3\lib\site-packages\numba\npyufunc\decorators.py", line 120, in wrap
    vec.add(sig)
  File "C:\Anaconda3\lib\site-packages\numba\npyufunc\dufunc.py", line 139, in add
    args, return_type = sigutils.normalize_signature(sig)
  File "C:\Anaconda3\lib\site-packages\numba\sigutils.py", line 35, in normalize_signature
    % (sig.__class__.__name__,))
TypeError: invalid signature: 'Float' instance not allowed

我可以通过在签名本身周围添加引号来修改签名，以便将其转换为字符串。当我这样做时，错误现在变成了这个：

  File "C:\home\git\indices_github\palmer.py", line 346, in <module>
    @numba.vectorize(['f8,f8,f8,f8(f8,f8,f8,f8,f8,f8,f8,f8)'])
  File "C:\home\Anaconda3\lib\site-packages\numba\npyufunc\decorators.py", line 120, in wrap
    vec.add(sig)
  File "C:\home\Anaconda3\lib\site-packages\numba\npyufunc\dufunc.py", line 139, in add
    args, return_type = sigutils.normalize_signature(sig)
  File "C:\home\Anaconda3\lib\site-packages\numba\sigutils.py", line 45, in normalize_signature
    check_type(ty)
  File "C:\home\Anaconda3\lib\site-packages\numba\sigutils.py", line 40, in check_type
    "got %r" % (ty,))
TypeError: invalid signature: expected a type instance, got (float64, float64, float64, float64, float64, float64, float64, float64) -> float64

有谁能在这里暗示我做错了什么？提前感谢任何建议。

用numba声明向量化的ufunc，签名错误：'Float'实例不被允许

0 个答案: