pandas 合并命令在并行循环中失败 - “ValueError：缓冲区源数组是只读的”答案

【问题标题】：pandas merge command failing in parallel loop - "ValueError: buffer source array is read-only"pandas 合并命令在并行循环中失败 - “ValueError：缓冲区源数组是只读的”
【发布时间】：2019-05-08 08:27:34
【问题描述】：

我正在使用并行循环和 pandas 编写引导算法。我遇到的问题是并行循环内的合并命令会导致“ValueError：缓冲区源数组是只读的”错误 - 但前提是我使用完整数据集进行合并（120k 行）。任何少于 12k 行的子集都可以正常工作，所以我推断这不是语法问题。我能做什么？

当前 pandas 版本为 0.24.2，cython 为 0.29.7。

_RemoteTraceback                          Traceback (most recent call last)
_RemoteTraceback: 
"""
Traceback (most recent call last):
  File "/home/ubuntu/.local/lib/python3.6/site-packages/joblib/externals/loky/process_executor.py", line 418, in _process_worker
    r = call_item()
  File "/home/ubuntu/.local/lib/python3.6/site-packages/joblib/externals/loky/process_executor.py", line 272, in __call__
    return self.fn(*self.args, **self.kwargs)
  File "/home/ubuntu/.local/lib/python3.6/site-packages/joblib/_parallel_backends.py", line 567, in __call__
    return self.func(*args, **kwargs)
  File "/home/ubuntu/.local/lib/python3.6/site-packages/joblib/parallel.py", line 225, in __call__
    for func, args, kwargs in self.items]
  File "/home/ubuntu/.local/lib/python3.6/site-packages/joblib/parallel.py", line 225, in <listcomp>
    for func, args, kwargs in self.items]
  File "<ipython-input-72-cdb83eaf594c>", line 12, in bootstrap
  File "/home/ubuntu/.local/lib/python3.6/site-packages/pandas/core/frame.py", line 6868, in merge
    copy=copy, indicator=indicator, validate=validate)
  File "/home/ubuntu/.local/lib/python3.6/site-packages/pandas/core/reshape/merge.py", line 48, in merge
    return op.get_result()
  File "/home/ubuntu/.local/lib/python3.6/site-packages/pandas/core/reshape/merge.py", line 546, in get_result
    join_index, left_indexer, right_indexer = self._get_join_info()
  File "/home/ubuntu/.local/lib/python3.6/site-packages/pandas/core/reshape/merge.py", line 756, in _get_join_info
    right_indexer) = self._get_join_indexers()
  File "/home/ubuntu/.local/lib/python3.6/site-packages/pandas/core/reshape/merge.py", line 735, in _get_join_indexers
    how=self.how)
  File "/home/ubuntu/.local/lib/python3.6/site-packages/pandas/core/reshape/merge.py", line 1130, in _get_join_indexers
    llab, rlab, shape = map(list, zip(* map(fkeys, left_keys, right_keys)))
  File "/home/ubuntu/.local/lib/python3.6/site-packages/pandas/core/reshape/merge.py", line 1662, in _factorize_keys
    rlab = rizer.factorize(rk)
  File "pandas/_libs/hashtable.pyx", line 111, in pandas._libs.hashtable.Int64Factorizer.factorize
  File "stringsource", line 653, in View.MemoryView.memoryview_cwrapper
  File "stringsource", line 348, in View.MemoryView.memoryview.__cinit__
ValueError: buffer source array is read-only
"""

The above exception was the direct cause of the following exception:

ValueError                                Traceback (most recent call last)
<ipython-input-73-652c1db5701b> in <module>()
      1 num_cores = multiprocessing.cpu_count()
----> 2 results = Parallel(n_jobs=num_cores, prefer='processes', verbose = 5)(delayed(bootstrap)() for i in range(n_trials))
      3 #pd.DataFrame(results[0])

~/.local/lib/python3.6/site-packages/joblib/parallel.py in __call__(self, iterable)
    932 
    933             with self._backend.retrieval_context():
--> 934                 self.retrieve()
    935             # Make sure that we get a last message telling us we are done
    936             elapsed_time = time.time() - self._start_time

~/.local/lib/python3.6/site-packages/joblib/parallel.py in retrieve(self)
    831             try:
    832                 if getattr(self._backend, 'supports_timeout', False):
--> 833                     self._output.extend(job.get(timeout=self.timeout))
    834                 else:
    835                     self._output.extend(job.get())

~/.local/lib/python3.6/site-packages/joblib/_parallel_backends.py in wrap_future_result(future, timeout)
    519         AsyncResults.get from multiprocessing."""
    520         try:
--> 521             return future.result(timeout=timeout)
    522         except LokyTimeoutError:
    523             raise TimeoutError()

/usr/lib/python3.6/concurrent/futures/_base.py in result(self, timeout)
    430                 raise CancelledError()
    431             elif self._state == FINISHED:
--> 432                 return self.__get_result()
    433             else:
    434                 raise TimeoutError()

/usr/lib/python3.6/concurrent/futures/_base.py in __get_result(self)
    382     def __get_result(self):
    383         if self._exception:
--> 384             raise self._exception
    385         else:
    386             return self._result

ValueError: buffer source array is read-only

代码是

def bootstrap():

    df_resample_ids = skl.utils.resample(ob_ids)
    df_resample_ids = pd.DataFrame(df_resample_ids).sort_values(by="0").reset_index(drop=True)
    df_resample_ids.columns = [ob_id_field]

    df_resample = pd.DataFrame(df_resample_ids.merge(df, on = ob_id_field))

    return df_resample

num_cores = multiprocessing.cpu_count()
results = Parallel(n_jobs=num_cores, prefer='processes', verbose = 5)(delayed(bootstrap)() for i in range(n_trials))

算法将从 ID 变量创建重新采样/替换的 ID，并使用合并命令根据重新采样的 ID 和存储在 df 中的原始数据集创建新数据集。如果我剪掉原始数据集的一个子集（任何地方），留下不到 12k 行，那么并行循环将完成而不会出现错误并按预期执行。

根据要求，下面是一个新的 sn-p，用于重新创建数据结构并反映我目前正在研究的主要方法：

import pandas as pd
import sklearn as skl
import multiprocessing
from joblib import Parallel, delayed

df = pd.DataFrame(np.random.randn(200000, 24), columns=list('ABCDDEFGHIJKLMNOPQRSTUVW'))
df["ID"] = df.index.drop_duplicates().tolist() 
ob_ids = df.index.drop_duplicates().tolist() 

def bootstrap2():

    df_resample_ids = skl.utils.resample(ob_ids)
    df_resample_ids = pd.DataFrame(df_resample_ids).sort_values(by=0).reset_index(drop=True)
    df_resample_ids.columns = ['ID']
    df_resample = pd.DataFrame(df1.merge(df_resample_ids, on = 'ID'))

    result = df_resample

    return result

num_cores = multiprocessing.cpu_count()
results = Parallel(n_jobs=num_cores, prefer='processes', verbose = 5)(delayed(bootstrap2)() for i in range(n_trials))

但是，我注意到当数据完全由 np.random 数字组成时，循环通过而没有错误。原始数据框的dtypes是：

start_rtg                        int64
end_rtg                        float64
days_diff                      float64
ultimate_customer_system_id      int64

如何避免只读错误？

【问题讨论】：

您可以编辑问题以使其独立吗？例如将代码添加到问题中以从随机数据生成数据框，确保列出所有导入并显示使用的函数来自何处。见stackoverflow.com/help/mcve
ok 添加信息以使其更加独立。谢谢！
这个问题是谷歌上关于这个错误的第一个结果，所以对于未来的读者来说，这显然是某些版本的 joblib 的错误，请参阅github.com/scikit-learn/scikit-learn/issues/7981。解决方法是在Parallel() 函数调用中添加max_nbytes=None。

标签： python python-3.x pandas multiprocess

【解决方案1】：

发布我的问题的答案，因为我发现其中一个变量是 int64 数据类型。当我将所有变量转换为 float64 时，错误消失了。所以这是一个仅限于某些数据类型的问题......

干杯斯蒂芬

【讨论】：