Why does Pandas raise a "ZeroDivisionError" in one case, but not in another?

Question

I have a Pandas dataframe 'dt = myfunc ()' and copy the screen output from IDLE as shown below:

>>> from __future__ import division >>> dt = __get_stk_data__(['*'], frq='CQQ', from_db=False) # my function >>> dt = dt[dt['ebt']==0][['tax','ebt']] >>> type(dt) <class 'pandas.core.frame.DataFrame'> >>> dt tax ebt STK_ID RPT_Date 000719 20100331 0 0 20100630 0 0 20100930 0 0 20110331 0 0 002164 20080331 0 0 300155 20120331 0 0 600094 20090331 0 0 20090630 0 0 20090930 0 0 600180 20090331 0 0 600757 20110331 0 0 >>> dt['tax_rate'] = dt.tax/dt.ebt Traceback (most recent call last): File "<stdin>", line 1, in <module> File "D:\Python\Lib\site-packages\pandas\core\series.py", line 72, in wrapper return Series(na_op(self.values, other.values), File "D:\Python\Lib\site-packages\pandas\core\series.py", line 53, in na_op result = op(x, y) ZeroDivisionError: float division >>> 

It costs me a lot of time to find out why Pandas raises "ZeroDivisionError: float division" and Pandas works fine for the code below the sample:

 tuples = [('000719','20100331'),('000719','20100930'),('002164','20080331')] index = MultiIndex.from_tuples(tuples, names=['STK_ID', 'RPT_Date']) dt =DataFrame({'tax':[0,0,0],'ebt':[0,0,0]},index=index) dt['tax_rate'] = dt.tax/dt.ebt >>> dt ebt tax tax_rate STK_ID RPT_Date 000719 20100331 0 0 NaN 20100930 0 0 NaN 002164 20080331 0 0 NaN >>> 

I expect Pandas to offer "NaN" for both cases, why does "ZeroDivisionError" happen in the first case? How to fix it?

---

below codes and output screen are connected to provide additional information for debugging

 def __by_Q__(df): ''' this function transforms the input financial report data (which is accumulative) to qurterly data ''' df_q1=df[df.index.map(lambda x: x[1].endswith("0331"))] print 'before diff:\n' print df.dtypes df_delta = df.diff() print '\nafter diff: \n' print df_delta.dtypes q1_mask = df_delta.index.map(lambda x: x[1].endswith("0331")); df_q234 = df_delta[~q1_mask] rst = concat([df_q1,df_q234]) rst=rst.sort_index() return rst 

screen output:

 before diff: sales float64 discount object net_sales float64 cogs float64 ebt float64 tax float64 after diff: sales object discount object net_sales object cogs object ebt object tax object 

Answer 1

@bigbug, how do you get data from the SQLite backend? If you look at pandas.io.sql , the read_frame method has a coerce_float parameter, which, if possible, should convert numerical data to a float.

The second example works because the DataFrame constructor is trying to be smart with respect to types. If you set the dtype object to an object, it will not execute:

 In [16]: dt = DataFrame({'tax':[0,0,0], 'ebt':[0,0,0]},index=index,dtype=object) In [17]: dt.tax/dt.ebt --------------------------------------------------------------------------- ZeroDivisionError Traceback (most recent call last) 

Check your data import code again and let me know what you found?

Answer 2

I was not able to reproduce this behavior (I tried to create DataFrames from integers, float and numpy arrays), buy, I think it is better to think of NaN in the tax_rate column, and then overwrite the values ​​when ebt not zero:

 dt['tax_rate'] = numpy.nan dt['tax_rate'][dt.ebt != 0] = dt.tax[dt.ebt != 0] / dt.ebt[dt.ebt != 0]