如何修复/调试 scikit learn 中引发的这个多进程终止工作错误

问题描述

我最近设置了一台新机器,以帮助减少拟合模型和数据处理的运行时间.

I recently set up a new machine to aid in decreasing run times for fitting models and data wrangling.

我做了一些初步的基准测试,一切都很顺利,但是当我尝试在 scikit learn 中启用多进程工作者时遇到了障碍.

I did some preliminary benchmarks and everything is mostly smoothe, but I ran into a snag when I tried enabling multi-process workers with in scikit learn.

我已将错误简化为与我的原始代码无关,因为我在不同的机器和 VM 上启用了此功能而没有问题.

I've simplified the error to not be associated with my original code as I enabled this feature without a problem on a different machine and a VM.

我还进行了内存分配检查,以确保我的机器没有用完可用的 RAM.我有 16gb 的 RAM,所以应该没有问题,但我留下了测试的输出,以防我错过了一些东西.

I've also done memory allocation checks to make sure my machine wasn't running out of available RAM. I have 16gb of RAM so there should be no issue, but I've left the output of the test incase I missed something.

鉴于附近的回溯错误,我可以告诉我的操作系统正在杀死它,但对于我的生活,我无法弄清楚为什么.据我所知,我的代码仅在仅使用单个 CPU 内核时才会运行.

Given the traceback error near I can tell my OS is killing this, but for the life of me I can't figure out why. Near as I can tell my code will ONLY run when it is just using a single CPU core.

我运行的是 Windows 10、AMD ryzen 7 2700x、16GB RAM

I'm running Windows 10, AMD ryzen 7 2700x, 16GB RAM

import sklearn
import numpy as np
import tracemalloc
import time


from sklearn.model_selection import cross_val_score
from numpy.random import randn
from sklearn.linear_model import Ridge


##################### memory allocation snapshot

tracemalloc.start()

start_time = time.time()
snapshot1 = tracemalloc.take_snapshot()

###################### model

X = randn(815000, 100)
y = randn(815000, 1)
mod = Ridge()
sc = cross_val_score(mod, X, y,verbose =10, n_jobs=3)

################### Second memory allocation snapshot

snapshot2 = tracemalloc.take_snapshot()
top_stats = snapshot2.compare_to(snapshot1, 'lineno')

print("[ Top 10 ]")
for stat in top_stats[:5]:
print(stat)

由此得出的预期结果非常明显,只是拟合模型的返回分数.

The expected results from this are pretty obvious, just a returned score with the fit model.

[Parallel(n_jobs=3)]: Using backend LokyBackend with 3 concurrent workers.
[Parallel(n_jobs=3)]: Done   3 out of   3 | elapsed:    0.2s remaining:    0.0s
---------------------------------------------------------------------------
TerminatedWorkerError                     Traceback (most recent call last)
<ipython-input-18-b2bdfd425f82> in <module>
     16 y = randn(815000, 1)
     17 mod = Ridge()
---> 18 sc = cross_val_score(mod, X, y,verbose =10, n_jobs=3)

..........

TerminatedWorkerError: A worker process managed by the executor was unexpectedly terminated. 
This could be caused by a segmentation fault while calling the function or by an excessive memory usage causing the Operating System to kill the worker.

内存输出

[ Top 5 ]
<ipython-input-18-b2bdfd425f82>:15: size=622 MiB (+622 MiB), count=3 (+3), average=207 MiB
<ipython-input-18-b2bdfd425f82>:16: size=6367 KiB (+6367 KiB), count=3 (+3), average=2122 KiB
~python37libinspect.py:732: size=37.2 KiB (+26.2 KiB), count=596 (+419), average=64 B
~python37libsite-packagessklearnexternalsjoblib
umpy_pickle.py:292: size=7072 B (+3808 B), count=13 (+7), average=544 B
~python37libpickle.py:549: size=5728 B (+3408 B), count=14 (+8), average=409 B


解决方案

我发现我的 scipy 模块与我的 windows 10 C++ 可再发行版本不兼容.

I figured out the my scipy module was incompatible with my windows 10 C++ redistributable version.

我所做的只是下载最新的 Visual Studio 并安装单个组件"部分中列出的 C++ 可再发行更新.

All i did was download the latest visual studio and installed the C++ redistributable update that is listed in the "individual components" section.

安装后,我重新启动计算机并运行.

Once I installed that I restarted my computer and ran.

import scipy
scipy.test()

一旦它实际运行,我尝试了上面的代码块并修复了它.

Once that was actually running I attempted my code block above and it fixed.

我认为这归结为使用全新版本的 python 和 scipy 安装旧版本的 windows 10

I think what this boils down to is installing an old build of windows 10 with a brand new version of python and scipy

这需要很长时间来解决和调试.希望对您有所帮助.

This took a LONG time to solve and debug. Hopefully it helps.

相关文章