TypeError when appending fields to a structured array of size ONE

问题

I'm getting a run-time error when trying to append field(s) to a structured array of size ONE. I've written a simple example below:

import numpy as np
import numpy.lib.recfunctions as rcfuncs

dtype_ = np.dtype( { 'names': ["field_a","field_b","field_c"]
                  , 'formats': ['S32', 'i4', 'f8']}
                  )
data_ = [("1",17, 123.45)]
numpy_array = np.array(data_, dtype_)            

# append 2 fields
numpy_array = rcfuncs.append_fields( numpy_array,["field_d","field_e"],data=[ "1","3" ] )

# append 1 field fails :(
numpy_array = rcfuncs.append_fields( numpy_array, "field_f", data=["123456"] )

I'm getting the error:

TypeError: descriptor 'ravel' requires a 'numpy.ndarray' object but received a 'numpy.void'

As well, if I 'invert' the appends, the statement with the two fields append will fail:

# append 1 field
numpy_array = rcfuncs.append_fields( numpy_array, "field_f", data=["123456"] )

# append 2 fields fails :(
numpy_array = rcfuncs.append_fields( numpy_array,["field_d","field_e"],data=[ "1", "3" ] )

I am running with python 2.7.11 and numpy 1.11.0 and I do not have the issue when the initial array is of size greater than 2.

How to solve the type error?

Thanks

回答1:

We do not get the TypeError when setting the optional parameter usemask to False

numpy_array = \
  rcfuncs.append_fields(numpy_array, "field_f", data=["123456"], usemask=False)
numpy_array = \
  rcfuncs.append_fields(numpy_array,["field_d","field_e"],data=[ "1", "3" ], usemask=False)

回答2:

For reference, here's the full traceback

Start with a structured array, with one record:

array([('1', 17, 123.45)], 
      dtype=[('field_a', 'S32'), ('field_b', '<i4'), ('field_c', '<f8')])

After the first append, we have a masked array, still with 1 record:

masked_array(data = [('1', 17, 123.45, '1', '3')],
             mask = [(False, False, False, False, False)],
       fill_value = ('N/A', 999999, 1e+20, 'N', 'N'),
            dtype = [('field_a', 'S32'), ('field_b', '<i4'), ('field_c', '<f8'), ('field_d', 'S1'), ('field_e', 'S1')])

The error looks like it has more to do with the masked array code than recfunctions - though I'll have to look at the code to see why it is using ravel.

Traceback (most recent call last):
  File "stack36440557.py", line 15, in <module>
    numpy_array2 = rcfuncs.append_fields( numpy_array1, "field_f", data=["123456"] ,usemask=False)
  File "/usr/local/lib/python2.7/site-packages/numpy/lib/recfunctions.py", line 633, in append_fields
    base = merge_arrays(base, usemask=usemask, fill_value=fill_value)
  File "/usr/local/lib/python2.7/site-packages/numpy/lib/recfunctions.py", line 389, in merge_arrays
    seqarrays = seqarrays.ravel()
  File "/usr/local/lib/python2.7/site-packages/numpy/ma/core.py", line 4022, in ravel
    r = ndarray.ravel(self._data).view(type(self))
TypeError: descriptor 'ravel' requires a 'numpy.ndarray' object but received a 'numpy.void'

So one fix is to turn off the use of masked_arrays. That isn't needed unless the added fields are missing some data.

Another is to put the 1st argument in a list:

rcfuncs.append_fields( [numpy_array1], "field_f", data=['12345'])

append_fields(base, ....) calls

merge_arrays(base, usemask=usemask, fill_value=fill_value)

which in turn calls

base.ravel()  # now call seq_arrays

But first it checks it is length one

# Only one item in the input sequence ?
if (len(seqarrays) == 1):
    seqarrays = np.asanyarray(seqarrays[0])

For a simple structured array, y, and its masked equivalent, ym:

In [405]: y
Out[405]: 
array([(b'xxx', 1)], 
      dtype=[('f0', 'S5'), ('f1', '<i4')])
In [406]: ym=np.ma.masked_array(y)

This length 1 action produces another array for the regular structured array:

In [407]: np.asanyarray(y[0])
Out[407]: 
array((b'xxx', 1), 
      dtype=[('f0', 'S5'), ('f1', '<i4')])

but a void (structured array record/element) for the masked one:

In [408]: np.asanyarray(ym[0])
Out[408]: (b'xxx', 1)
In [409]: type(np.asanyarray(ym[0]))
Out[409]: numpy.ma.core.mvoid

np.asanyarray(ym[0]).ravel() produces this TypeError.

If the base is a list, [ym], this just extracts ym. If the base is (2,) or longer, it doesn't pass though this statement.

I haven't thought of fix yet - other the user level kludge of passing the masked array in a list.

A possible fix is to simply remove this base=merge_arrays(base...) line (in append_fields). But I need to know why it is there in the first place. The intent may be to cleanup certain base array inputs.

The unit test file, test/test_recfunctions.py runs fine with this line commented out.

I've added a comment on this to an old numpy issue

https://github.com/numpy/numpy/issues/2346

来源：https://stackoverflow.com/questions/36440557/typeerror-when-appending-fields-to-a-structured-array-of-size-one

标签

python

numpy

structured-array