英文:
Weird behavior while using numba on a raspberry pi zero
问题
我在树莓派 Zero 上使用 numba 时遇到了一些奇怪的行为。当我使用浮点数时，我的函数返回了数学上错误的奇怪计算结果。这是我的代码：
import numba
@numba.jit(nopython=True)
def numtesting(n):
print(n)
print(n)
numtesting(1.0)
我期望的输出是：
- 1.0
- 1.0
但实际得到的是：
- 1.0
- 0.031326431754716
有什么想法发生了什么？
英文:
I got some weird behavior while using numba on a raspberry pi zero. When I use floating numbers, my function returns weird calculations which are mathematically wrong. This is my code:
import numba
@numba.jit(nopython=True)
def numtesting(n):
print(n)
print(n)
numtesting(1.0)
I would expect
- 1.0
- 1.0
Instead i get:
- 1.0
- 0.031326431754716
Any ideas what's happening?
答案1
得分: 0
我放弃在我的树莓派 Zero 上安装 numba。在 GitHub 上找到的信息显示，numba 只在 64 位架构上正常工作。不过，我购买了一台带有 64 位架构的树莓派 Zero 2，并且成功在上面安装了 numba 0.56.4。然而，我不得不使用一些变通方法，以下是我是如何做到的：
操作系统：树莓派 OS Lite（32 位）
为了成功安装 llvmlite，我执行了 sudo apt install -y llvm-dev libffi-dev
为了使 numpy 正常运行，我执行了 sudo apt-get install -y libatlas libopenjp2-7
之后，我在虚拟环境中使用了 pip install llvmlite==0.39.1
在这一步，我遇到了一个问题，安装程序在为 llvmlite 构建 wheel 时卡住了。ChatGPT 建议我将 /etc/dphys-swapfile 中的 SWAP 大小增加到 1024（默认值为 100）。参数名为 CONF_SWAPSIZE
之后，我能够使用 pip install numpy==0.56.4 将 numpy 安装到我的虚拟环境中
希望这对某人有所帮助。我花了很多时间才解决了这个问题。
英文:
I gave up installing numba on my Raspberry Pi Zero. Found infos on github that numba only correctly works on 64-bit architectures. However, I bought myself a Raspberry Pi Zero 2 with 64-bit architecture and I managed to install numba 0.56.4 on it. However, I had to use some workarounds, here is how i did it:
OS: Raspberry Pi OS Lite (32bit)
For llvmlite to be installed successfully, I installed sudo apt install -y llvm-dev libffi-dev
For numpy to run successfully, I installed sudo apt-get install -y libatlas libopenjp2-7
After that i used pip install llvmlite==0.39.1 into a virtual environment
At this point I faced the issue, that the installer froze builing a wheel for llvmlite. ChatGPT helped by recommending to increase the SWAP size in /etc/dphys-swapfile to 1024 (Default was 100). The parameter is called CONF_SWAPSIZE
After that i was able to install numpy into my virtual environment using pip install numpy==0.56.4
I really hope it helps someone. I took me quite some time to figure this out...
通过集体智慧和协作来改善编程学习和解决问题的方式。致力于成为全球开发者共同参与的知识库，让每个人都能够通过互相帮助和分享经验来进步。
评论