dask數據框中的.join結果似乎取決於方式，生成了dask數據框

Question

將join應用於由.from_delayed方法生成的dask數據框時，我得到了意外的結果。我想通過下面的例子來演示這個例子，它由三部分組成。

1. 生成經由from_delayed方法DASK數據幀，並通過from_pandas
2. 產生的數據幀DASK加入它轉換既dataframes到大熊貓與compute方法dataframes。 （1）
3. 將通過from_delayed方法生成的dask數據幀轉換爲使用compute的pandas。使用from_pandas將其轉換回dask。然後加入（1）。

考慮下面的代碼：

import dask.dataframe 
import pandas as pd 

# functions for generating a dask dataframe 
def get_pdf(character): 
    '''constructs a pandas dataframe with indexes [character]1, ..., [character]5''' 
    index = [character + str(i) for i in range(5)] 
    return pd.DataFrame({'A':[1,2,3,4,5]}, index = index) 

def get_ddf(): 
    '''constructs dask dataframe out of pandas dataframes via the .from-delayed method with indexes A1, A2, A3, ... F3, F3, F4''' 
    delayed_list = [dask.delayed(get_pdf)(x) for x in 'ABCDEF'] 
    return dask.dataframe.from_delayed(delayed_list) 

#generate dask dataframes, that will be joined 
ddf1 = get_ddf() 
ddf2 = dask.dataframe.from_pandas(pd.DataFrame({'B': [1,2,3]}, index = ['A0', 'B1', 'C3']), npartitions = 2) 

#recreate ddf1 by converting it to a pandas dataframe and afterwards to a dask dataframe 
ddf1_from_pandas = dask.dataframe.from_pandas(ddf1.compute(), npartitions = 3) 

#compute joins 
dask_from_delayed_join = ddf1.join(ddf2, how = 'inner') 
pandas_join = ddf1.compute().join(ddf2.compute(), how = 'inner') 
dask_from_pandas_join = ddf1_from_pandas.join(ddf2, how = 'inner') 

我希望所有的三個結果（dask_from_delayed_join，pandas_join，dask_from_pandas_join）是相同的。

然而，第一結果不同於其他：

print(dask_from_delayed_join.compute())：

Empty DataFrame 
Columns: [A, B] 
Index: [] 

print(pandas_join)：

A B 
A0 1 1 
B1 2 2 
C3 4 3 

print(dask_from_pandas_join.compute())：

A B 
A0 1 1 
B1 2 2 
C3 4 3 

發生了什麼事？

Answer 1

dd.merge確實存在一些問題。這些已在dask版本0.10.2解決

In [10]: print(dask_from_delayed_join.compute()) 
    A B 
A0 1 1 
B1 2 2 
C3 4 3 

In [11]: print(pandas_join) 
    A B 
A0 1 1 
B1 2 2 
C3 4 3 

In [12]: print(dask_from_pandas_join.compute()) 
    A B 
A0 1 1 
B1 2 2 
C3 4 3