python 中的 cupy 库中的 amax 和 max 函数在使用只有一列或只有一行的矩阵时是否会出错？

百变鹏仔 1天前 #Python

文章标签矩阵

问题内容

我尝试使用 cupy 进行 gpu 加速来实现用于机器学习和图像分类的 softmax 激活函数。我观察到，对于形状为 nx1 或 1xn 的数组，cupys max 函数会输出错误。然而，对于 nxa 的所有其他情况（其中 n 和 a 都是 1 以外的整数），它工作得很好。

我的代码：

def softmax_(z):    max_z = cp.max(z, axis=0, keepdims=true)  # problematic max function    exp_z = cp.exp(z - max_z)  # subtracting the maximum value for numerical stability    sum_exp_z = cp.sum(exp_z, axis=0, keepdims=true)  # summing up the values    return exp_z / sum_exp_z  # dividing them to get the softmaxarray1 = cp.random.randn(3, 4)  # 3x4array2 = cp.random.randn(5, 1)  # 5x1print(softmax_(array1))  # no errorprint(softmax_(array2))  # produces an error

我的操作系统错误，我对此缺乏经验：

oserror: [winerror 123] the filename, directory name, or volume label syntax is incorrect: 'c:\users\confidential\.cupy\jitify_cache\tmp1pxgjv_g' -&gt; 'c:\users\confidential/.cupy/jitify_cache/jitify_<unknown>_200200_12030_2_b3452ffa79e273adadd0403b6b0c05b78158b1e0.json'</unknown>

数组 1 的输出

立即学习“Python免费学习笔记（深入）”；

output:  [[0.17813469 0.20912114 0.19734889 0.30515635]  [0.425690720.47354802 0.4463671  0.20997539]  [0.39617459 0.31733085 0.356284   0.48486825]]

数组2的错误：

../../util_ptx.cuh(38): warning: util_type.cuh: [jitify] File not found ../../util_ptx.cuh(41): warning: util_debug.cuh: [jitify] File not found../../thread/thread_load.cuh(40): warning: ../util_ptx.cuh: [jitify] File not foundTraceback (most recent call last):  File "c:UsersconfidentialDesktopProjekteNeural_network2	est.py", line 14, in <module>            print(softmax_(array2))          ^^^^^^^^^^^^^^^^  File "c:UsersconfidentialDesktopProjekteNeural_network2	est.py", line 4, in softmax_    `max_Z = cp.max(Z, axis=0, keepdims=True)`            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^  File "C:UsersconfidentialPycharmProjectsunpyenvLibsite-packagescupy_statisticsorder.py", line 81, in amax    return a.max(axis=axis, out=out, keepdims=keepdims)           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^  File "cupy_corecore.pyx", line 990, in cupy._core.core._ndarray_base.max  File "cupy_corecore.pyx", line 998, in cupy._core.core._ndarray_base.max  File "cupy_core_routines_statistics.pyx", line 43, in cupy._core._routines_statistics._ndarray_max  File "cupy_core_reduction.pyx", line 618, in cupy._core._reduction._SimpleReductionKernel.__call__  File "cupy_core_reduction.pyx", line 370, in cupy._core._reduction._AbstractReductionKernel._call  File "cupy_core_cub_reduction.pyx", line 689, in cupy._core._cub_reduction._try_to_call_cub_reduction  File "cupy_core_cub_reduction.pyx", line 540, in cupy._core._cub_reduction._launch_cub      File "cupy_util.pyx", line 64, in cupy._util.memoize.decorator.ret  File "cupy_core_cub_reduction.pyx", line 240, in cupy._core._cub_reduction._SimpleCubReductionKernel_get_cached_function  File "cupy_core_cub_reduction.pyx", line 223, in cupy._core._cub_reduction._create_cub_reduction_function  File "cupy_corecore.pyx", line 2254, in cupy._core.core.compile_with_cache  File "C:UsersconfidentialPycharmProjectsunpyenvLibsite-packagescupycudacompiler.py", line 484, in _compile_module_with_cache    return _compile_with_cache_cuda(           ^^^^^^^^^^^^^^^^^^^^^^^^^  File "C:UsersconfidentialPycharmProjectsunpyenvLibsite-packagescupycudacompiler.py", line 562, in _compile_with_cache_cuda    ptx, mapping = compile_using_nvrtc(                   ^^^^^^^^^^^^^^^^^^^^  File "C:UsersconfidentialPycharmProjectsunpyenvLibsite-packagescupycudacompiler.py", line 319, in compile_using_nvrtc    return _compile(source, options, cu_path,           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^  File "C:UsersconfidentialPycharmProjectsunpyenvLibsite-packagescupycudacompiler.py", line 284, in _compile    options, headers, include_names = _jitify_prep(                                      ^^^^^^^^^^^^^  File "C:UsersconfidentialPycharmProjectsunpyenvLibsite-packagescupycudacompiler.py", line 233, in _jitify_prep    jitify._init_module()  File "cupycudajitify.pyx", line 212, in cupy.cuda.jitify._init_module  File "cupycudajitify.pyx", line 233, in cupy.cuda.jitify._init_module  File "cupycudajitify.pyx", line 209, in cupy.cuda.jitify._init_cupy_headers  File "cupycudajitify.pyx", line 198, in cupy.cuda.jitify._init_cupy_headers_from_scratch    File "cupycudajitify.pyx", line 128, in cupy.cuda.jitify.dump_cacheOSError: [WinError 123] The syntax for the file name, directory name, or volume label is incorrect: 'C:\Users\confidential\.cupy\jitify_cache\tmps16uxq46' -&gt; 'C:\Users\confidential/.cupy/jitify_cache/jitify_<unknown>_200200_12030_2_b3452ffa79e273adadd0403b6b0c05b78158b1e0.json'</unknown></module>

正确答案

您需要遵循的一些调试步骤。

1）更新cupy

pip install cupy --upgrade

2) 检查权限。确保运行脚本的用户具有读取和写入 cupy_cache_dir 环境变量中指定的缓存目录的必要权限。

重塑输入数组如果问题仍然存在，您可以尝试将输入数组重塑为 '(n,)' 的形状，而不是 '(n, 1)' 或 '(1, n)'。

4)禁用jit编译您可以尝试通过将 cupy_cache_dir 环境变量设置为有效目录来禁用 jit 编译。

import cupy as cpimport osos.environ['CUPY_CACHE_DIR'] = '/path/to/valid/directory'

将“/path/to/valid/directory”替换为 cupy 可以成功缓存已编译内核的目录。这可能会帮助您避免 oserror。

文章推荐

python 中的 cupy 库中的 amax 和 max 函数在使用只有一列或只有一行的矩阵时是否会出错？

正确答案

为何Python编程是当前最佳的就业方向？

Python程序员的就业前景及发展路径

探索Python编程领域中最热门的就业选项

在Python中，for循环和while循环之间的区别是什么？

什么样的企业最需要Python开发人员？