我目前正在尝试使用带有多处理后端的joblib库在python 3.5中运行并行进程。但是，每次运行时都会出现此错误：
Process ForkServerPoolWorker-5:
Traceback (most recent call last):
File “/opt/anaconda3/lib/python3.5/multiprocessing/process.py”, line 249, in _bootstrap
self.run()
File “/opt/anaconda3/lib/python3.5/multiprocessing/process.py”, line 93, in run
self._target(*self._args, **self._kwargs)
File “/opt/anaconda3/lib/python3.5/multiprocessing/pool.py”, line 108, in worker
task = get()
File “/opt/anaconda3/lib/python3.5/site-packages/joblib/pool.py”, line 362, in get
return recv()
File “/opt/anaconda3/lib/python3.5/multiprocessing/connection.py”, line 251, in recv
return ForkingPickler.loads(buf.getbuffer())
TypeError: __new__() missing 1 required positional argument: ‘path’这里是我用来运行它的joblib代码：
from joblib import Parallel, delayed
results = Parallel(n_jobs=6) (
delayed(func)(i) for i in array)默认情况下后端是多处理。当我将后端更改为“线程化”时，代码运行正常，但对于我的用例，与使用多处理相比，线程化效率低下。
我也尝试直接使用下面的代码使用多处理，但我仍然得到相同的错误：
from multiprocessing import Pool
with Pool(5) as p:
results = p.map(func, array)编辑：这是我想如何使用joblib的一个较小的例子。我的实现具有包装在类函数中的joblib函数。下面的代码工作，但是当我用我的实际数据运行相同的功能时，我得到了上述错误。如果使用大量数据，以下功能是否会导致问题？
import numpy as np
import pandas as pd
from joblib import Parallel, delayed
import multiprocessing
class ABTest(object):
def __init__(self, df, test_comps, test_name):
self.df = df
self.test_comps = test_comps
self.test_name = test_name
self.variant_dict = {}
for variant in self.df[‘test_variant’].unique():
self.variant_dict[variant] = self.df[self.df[‘test_variant’] == variant]
self.test_cols = [‘clicks’, ‘opens’, ‘buys’]
self.cpu_count = multiprocessing.cpu_count()
def bootstrap_ci(self, arguments):
”’
Finds the confidence interval for the difference in means for
two test groups using bootstrap
In: self
arugments (list) – A list with elements [test_comp, cols], where test_comp
is a tuple of two test variants and cols is a list
of columns to bootstrap means for. A single argument
must be used for parallel processing
Out: results (matrix) – confidence interval information for the difference
in means of the two groups
Creates: None
Modifies: None
”’
test_comp = arguments[0]
cols = arguments[1]
test_a_df = self.variant_dict[test_comp[0]]
test_b_df = self.variant_dict[test_comp[1]]
results = []
print(‘Getting Confidence Intervals for Test Groups: {}, {}…’.format(test_comp[0], test_comp[1]))
for col in cols:
test_a_sample_mean = []
test_b_sample_mean = []
test_a_len = test_a_df.shape[0]
test_b_len = test_b_df.shape[0]
for j in range(5000):
# Get sample means for both test variants
test_a_bs_mean = test_a_df[col].sample(n=test_a_len, replace=True).mean()
test_a_sample_mean.append(test_a_bs_mean)
test_b_bs_mean = test_b_df[col].sample(n=test_b_len, replace=True).mean()
test_b_sample_mean.append(test_b_bs_mean)
test_a_s = pd.Series(test_a_sample_mean)
test_b_s = pd.Series(test_b_sample_mean)
# Gets confidence interval for the difference in distribution of means
test_diffs = test_b_s-test_a_s
z = test_diffs.quantile([.025, 0.05, 0.5, 0.95, 0.975])
results.append([self.test_name, test_comp[0], test_comp[1], col, z.iloc[0], z.iloc[1], z.iloc[2], z.iloc[3], z.iloc[4]])
return results
def run_parallel(self, func, array):
”’
Runs a function (func) on each item in array and returns the results
In:
func (function that takes one argument) – the function to run in parallel
array (list or array-like object) – the array to iterate over
Out: results (list) – The results of running each item in array through func
Creates: None
Modifies: None
”’
# Never uses more than 6 cores
n_jobs = min(self.cpu_count – 1, 6)
results = Parallel(n_jobs=n_jobs) ( \
delayed(func) (i) for i in array)
return results
def confidence_intervals(self):
results = self.run_parallel(self.bootstrap_ci, [(x, self.test_cols) for x in self.test_comps])
results = np.array([y for x in results for y in x])
return results
if __name__ == ‘__main__’:
columns = [‘id’, ‘test_variant’, ‘clicks’, ‘opens’, ‘buys’]
data = [[0, ‘control’, 10, 60, 2], \
[1, ‘test_1’, 5, 50, 1], \
[2, ‘test_2’, 11, 50, 3], \
[3, ‘control’, 8, 55, 1], \
[4, ‘test_1’, 5, 40, 0], \
[5, ‘test_2’, 15, 100, 5], \
[6, ‘control’, 2, 30, 0], \
[7, ‘test_1’, 1, 60, 1], \
[8, ‘test_2’, 11, 50, 3], \
[9, ‘control’, 10, 60, 2], \
[10, ‘test_1’, 5, 50, 1], \
[11, ‘test_2’, 11, 50, 3], \
[12, ‘control’, 10, 60, 2], \
[13, ‘test_1’, 5, 50, 1], \
[14, ‘test_2’, 11, 50, 3], \
[15, ‘control’, 10, 60, 2], \
[16, ‘test_1’, 5, 50, 1], \
[17, ‘test_2’, 11, 50, 3], \
[18, ‘control’, 10, 60, 2], \
[19, ‘test_1’, 5, 50, 1], \
[20, ‘test_2’, 11, 50, 3]]
df = pd.DataFrame(data, columns=columns)
test_comps = [[‘control’, ‘test_1’], [‘control’, ‘test_2’], [‘test_1’, ‘test_2’]]
ab = ABTest(df, test_comps, test_name=’test’)
results = ab.confidence_intervals()
print(results)