我怎样才能让这个函数以数组作为输入，用 numba 编译？

Question

函数的签名是

def SLBQP(Q, q, u, a, x, eps=1e-6, maxIter=1000):

它返回一个 float64。

参数的类型是：

Q -- np.array([[1., 2.], [4., 5.]])
q -- np.array([1.,2.,3.,4.])
u -- a scalar
a -- np.array([1.,2.,3.,4.])
x -- np.array([1.,2.,3.,4.])

我试过

@jit('f8(f8[:,:], f8[:], f8, f8[:], f8[:], f8, i4)',nopython=True)
def SLBQP(Q, q, u, a, x, eps=1e-6, maxIter=1000):

它给了我这个错误：

Invalid use of Function(<built-in function array>) with argument(s) of type(s): (array(float64, 1d, C))
 * parameterized
In definition 0:
    TypingError: array(float64, 1d, C) not allowed in a homogeneous sequence
    raised from /Users/gerardozinno/Desktop/ProgettoML/venv/lib/python3.8/site-packages/numba/typing/npydecl.py:472
In definition 1:
    TypingError: array(float64, 1d, C) not allowed in a homogeneous sequence
    raised from /Users/gerardozinno/Desktop/ProgettoML/venv/lib/python3.8/site-packages/numba/typing/npydecl.py:472

我也试过这个：

@jit('numba.float64(numba.array(float64, 2d, C), numba.array(float64, 1d, C), numba.float64, numba.array(float64, 1d, C), numba.array(float64, 1d, C), numba.float64, numba.int64)',nopython=True)

它给了我一个语法错误。

编辑：

我试过这个签名：

@nb.njit('f8(f8[:,:], f8[:], f8, f8[:], f8[:], f8, i4)')

由 Thane Brooker 在答案部分建议，它给了我这个错误：

Invalid use of Function(<built-in function array>) with argument(s) of type(s): (array(float64, 1d, C))
 * parameterized
In definition 0:
    TypingError: array(float64, 1d, C) not allowed in a homogeneous sequence
    raised from /Users/gerardozinno/Desktop/ProgettoML/venv/lib/python3.8/site-packages/numba/typing/npydecl.py:472
In definition 1:
    TypingError: array(float64, 1d, C) not allowed in a homogeneous sequence
    raised from /Users/gerardozinno/Desktop/ProgettoML/venv/lib/python3.8/site-packages/numba/typing/npydecl.py:472
This error is usually caused by passing an argument of a type that is unsupported by the named function.
[1] During: resolving callee type: Function(<built-in function array>)
[2] During: typing of call at /Users/gerardozinno/Desktop/NUOVO/ProgettoML/svr/SLBQP.py (119)


File "SLBQP.py", line 119:
def SLBQP(Q, q, u, a, x, eps=1e-6, maxIter=1000):
    <source elided>
        v = np.dot(Qx,x) + np.dot(q, x)
        g = np.array(Qx+q)
        ^

Answer 1

这工作没有错误。

import numba as nb
import numpy as np

@nb.njit('f8(f8[:,:], f8[:], f8, f8[:], f8[:], f8, i4)')
def SLBQP(Q, q, u, a, x, eps=1e-6, maxIter=1000):
    return 1.

Q = np.array([[1., 2.], [4., 5.]])
q = np.array([1.,2.,3.,4.])
u = 50
a = np.array([1.,2.,3.,4.])
x = np.array([1.,2.,3.,4.])

result = SLBQP(Q, q, u, a, x, eps=1e-6, maxIter=1000)

我更改了您的示例 Q 变量（我认为这是一个错字），但否则我无法复制您的语法错误。 我猜你传递给函数的 Q 是一维数组，而不是你认为的二维数组。 查看Q.shape和Q.flags进行检查。

Answer 2

我通过在我的函数之前编写这段代码解决了这个问题。

from numba.extending import overload

@overload(np.array)
def np_array_ol(x):
    if isinstance(x, types.Array):
        def impl(x):
            return np.copy(x)
    return impl

@nb.njit('f8(f8[:,:], f8[:], f8, f8[:], f8[:], f8, i4)')
def SLBQP(Q, q, u, a, x, eps=1e-6, maxIter=1000):
    ...

显然，编辑部分中写入的错误是由 numba 内部的错误引起的。