隐藏Python模块中的中间计算(Hiding intermediate computations in Python module)

 我在Jupyter笔记本src/data.py有一个Python文件，它用于读取数据文件并提供一些输出。 
 
import numpy as np
import pandas as pd

sha256_perf = (
    pd.read_csv('data/hashbench-output.txt', sep='\t', na_filter=False)
        .query('Algorithm == "SHA256"')
)

mean_throughput = sha256_perf['Throughput (MiB/s)'].mean()
variance = sha256_perf['Error (± MiB/s)'] ** 2
total_variance = variance.sum()
row_count = sha256_perf.shape[0]
variance_of_mean = total_variance / (row_count ** 2)
error_of_mean = variance_of_mean ** 0.5

sha256_summary = pd.DataFrame(data=[[mean_throughput, error_of_mean]])
sha256_summary.columns = ['Mean Throughput (MiB/s)', 'Error (± MiB/s)']
 
 其中，我关心的唯一变量是输出表 - sha256_perf和sha256_summary 。 然而，Python无法知道，所以如果我dir()模块，我会得到一切： 
 
>>> import src.data as data
>>> dir(data)
['__builtins__', '__cached__', '__doc__', '__file__', '__loader__',
'__name__', '__package__', '__spec__', 'assumptions', 'error_of_mean', 
'mean_throughput', 'np', 'pd', 'prd_scratch_2018', 'row_count', 'sha256_perf', 
'sha256_summary', 'total_variance', 'util', 'variance', 'variance_of_mean']
 
 如果这是Ruby或Scala，我可以在块中初始化sha256_summary ，如下所示： 
 
sha256_summary = begin
  mean_throughput = sha256_perf['Throughput (MiB/s)'].mean()
  # ... etc. ...
  df.columns = ['Mean Throughput (MiB/s)', 'Error (± MiB/s)']
  df
end
 
 即使在Java（8+）中，我也可以与Supplier和lambda一起破解。 
 
 但据我所知，Python没有匿名块或多行lambda。 到目前为止，我所能做到的最好的一切就是把所有的东西放在一个函数中： 
 
def create_summary():
    mean_throughput = sha256_perf['Throughput (MiB/s)'].mean()
    # ... etc. ...
    sha256_summary.columns = ['Mean Throughput (MiB/s)', 'Error (± MiB/s)']
    return sha256_summary

sha256_summary = create_summary()
 
 但这仍然导出create_summary符号，我宁愿避免： 
 
>>> dir(data)
['__builtins__', '__cached__', '__doc__', '__file__', '__loader__', 
'__name__', '__package__', '__spec__', 'assumptions', 'create_summary', 
'np', 'pd', 'prd_scratch_2018', 'sha256_perf', 'sha256_summary', 'util']
 
 Pythonic避免污染全球命名空间的方式是什么？ 

I have a Python file in a Jupyter notebook, src/data.py, that's meant to read a data file and make some outputs available.
 
import numpy as np
import pandas as pd

sha256_perf = (
    pd.read_csv('data/hashbench-output.txt', sep='\t', na_filter=False)
        .query('Algorithm == "SHA256"')
)

mean_throughput = sha256_perf['Throughput (MiB/s)'].mean()
variance = sha256_perf['Error (± MiB/s)'] ** 2
total_variance = variance.sum()
row_count = sha256_perf.shape[0]
variance_of_mean = total_variance / (row_count ** 2)
error_of_mean = variance_of_mean ** 0.5

sha256_summary = pd.DataFrame(data=[[mean_throughput, error_of_mean]])
sha256_summary.columns = ['Mean Throughput (MiB/s)', 'Error (± MiB/s)']
 
Of this, the only variables I care about are the output tables -- sha256_perf and sha256_summary. However, Python has no way of knowing that, so if I dir() the module, I get everything:
 
>>> import src.data as data
>>> dir(data)
['__builtins__', '__cached__', '__doc__', '__file__', '__loader__',
'__name__', '__package__', '__spec__', 'assumptions', 'error_of_mean', 
'mean_throughput', 'np', 'pd', 'prd_scratch_2018', 'row_count', 'sha256_perf', 
'sha256_summary', 'total_variance', 'util', 'variance', 'variance_of_mean']
 
If this was Ruby or Scala, I could initialize sha256_summary in a block, something like:
 
sha256_summary = begin
  mean_throughput = sha256_perf['Throughput (MiB/s)'].mean()
  # ... etc. ...
  df.columns = ['Mean Throughput (MiB/s)', 'Error (± MiB/s)']
  df
end
 
Even in Java (8+), I could hack something together with a Supplier and a lambda.
 
But as far as I can tell, Python doesn't have anonymous blocks or multiline lambdas. So so far, the best I've been able to come up with is putting everything in a function:
 
def create_summary():
    mean_throughput = sha256_perf['Throughput (MiB/s)'].mean()
    # ... etc. ...
    sha256_summary.columns = ['Mean Throughput (MiB/s)', 'Error (± MiB/s)']
    return sha256_summary

sha256_summary = create_summary()
 
But this still exports the create_summary symbol, which I'd rather avoid:
 
>>> dir(data)
['__builtins__', '__cached__', '__doc__', '__file__', '__loader__', 
'__name__', '__package__', '__spec__', 'assumptions', 'create_summary', 
'np', 'pd', 'prd_scratch_2018', 'sha256_perf', 'sha256_summary', 'util']
 
What's the Pythonic way to avoid polluting the global namespace?

原文：https://stackoverflow.com/questions/49459373