I'm trying to make a an array of mixed types with an int and string, but am getting an error when populating the string.
a = np.full(2, np.nan, dtype='u4,S10'
a[0] = 1
a[1] = b'abc'
And that gives me the error
ValueError: invalid literal for int() with base 10: b'abc'
Is there a way to have an array of mixed types that I can put byte arrays into like this?
You are trying to assign to a whole row.
Try:
a = np.full(2, np.nan, dtype='u4, S10')
a[0][0] = 1
a[0][1] = 'abc'
In [246]: a = np.full(2, np.nan, dtype='u4,S10')
Look at the resulting array. Note the () making the 2 records. Also note that np.nan (a float) has been converted to other values according to the specified dtype (e.g. string 'nan'):
In [247]: a
Out[247]: array([(0, b'nan'), (0, b'nan')], dtype=[('f0', '<u4'), ('f1', 'S10')])
Assigning 1 - modifying both elements of the 1st record. Again, note the string assignment:
In [248]: a[0] = 1
In [249]: a
Out[249]: array([(1, b'1'), (0, b'nan')], dtype=[('f0', '<u4'), ('f1', 'S10')])
Access by field name:
In [250]: a['f0']
Out[250]: array([1, 0], dtype=uint32)
In [251]: a['f1']
Out[251]: array([b'1', b'nan'], dtype='|S10')
Modifying an element of the string field:
In [252]: a['f1'][1] = b'abc'
In [253]: a
Out[253]: array([(1, b'1'), (0, b'abc')], dtype=[('f0', '<u4'), ('f1', 'S10')])
Modifying both elements of a record - with a tuple:
In [254]: a[0] = (23, b'foobar')
In [255]: a
Out[255]:
array([(23, b'foobar'), ( 0, b'abc')],
dtype=[('f0', '<u4'), ('f1', 'S10')])
Initializing a structured array with a list of tuples:
In [256]: b = np.array([(2,b'xxx'),(34,b'xyz')], dtype=a.dtype)
In [257]: b
Out[257]: array([( 2, b'xxx'), (34, b'xyz')], dtype=[('f0', '<u4'), ('f1', 'S10')])
All of this is documented:
https://docs.scipy.org/doc/numpy/user/basics.rec.html#assigning-data-to-a-structured-array
