Weird error on int64 column about feather

[dan-tee]

Should add that converting the column from np.int64 to np.int32 fixed the problem. Then it saved just fine.

[wesm]

Can you let me know some more detail about your platform? The issue is that depending on your platform, NPY_INT64 may be either NPY_LONG or NPY_LONGLONG, and there may be code which sets NPY_LONGLONG (type=9) in spite of this. Easy to fix, but I'll have to look into a way to reproduce the error in a test case

[dan-tee]

I'm using a 2015 retina MacBookPro with OS X 10.11. Thanks for looking into it.

[jd155]

For what it's worth I got the same error when trying to save the following DataFrame. I have exactly the same OSX as dan-tee. Changing the columns to 'np.int32' also worked.

Data columns (total 5 columns):
Name Changed         433 non-null object
Name Changed         433 non-null int64
Name Changed         433 non-null int64
Name Changed         433 non-null int64
Name Changed         433 non-null int64
dtypes: int64(4), object(1)
memory usage: 20.3+ KB 

---

FeatherError                              Traceback (most recent call last)
<ipython-input-716-c3919d5846a7> in <module>()
      2 
      3 path = 'headcount.feather'
----> 4 feather.write_dataframe(Headcount, path)

/Users/Davies/anaconda/lib/python2.7/site-packages/feather/api.pyc in write_dataframe(df, path)
     33             name = str(name)
     34 
---> 35         writer.write_array(name, col)
     36 
     37     writer.close()

/Users/Davies/anaconda/lib/python2.7/site-packages/feather/ext.pyx in feather.ext.FeatherWriter.write_array (feather/ext.cpp:2130)()
     86             self.write_timestamp(name, col, mask)
     87         else:
---> 88             self.write_primitive(name, col, mask)
     89 
     90     cdef write_category(self, name, col, mask):

/Users/Davies/anaconda/lib/python2.7/site-packages/feather/ext.pyx in feather.ext.FeatherWriter.write_primitive (feather/ext.cpp:2435)()
    108 
    109         col_values = _unbox_series(col)
--> 110         self.write_ndarray(col_values, mask, &values)
    111         check_status(self.writer.get().AppendPlain(c_name, values))
    112 

/Users/Davies/anaconda/lib/python2.7/site-packages/feather/ext.pyx in feather.ext.FeatherWriter.write_ndarray (feather/ext.cpp:2742)()
    133     cdef int write_ndarray(self, values, mask, PrimitiveArray* out) except -1:
    134         if mask is None:
--> 135             check_status(pandas_to_primitive(values, out))
    136         else:
    137             check_status(pandas_masked_to_primitive(values, mask, out))

/Users/Davies/anaconda/lib/python2.7/site-packages/feather/ext.pyx in feather.ext.check_status (feather/ext.cpp:1546)()
     52 
     53     cdef string c_message = status.ToString()
---> 54     raise FeatherError(frombytes(c_message))
     55 
     56 set_numpy_nan(np.nan)

FeatherError: Invalid: unsupported type 9

[wesm]

I was able to reproduce the issue and fix it. Will push out a new PyPI release as soon as the Travis build finishes

[jd155]

Thanks. I'm using feather multiple times a day now. Really grateful for it!