MultiIndex.from_product throws on read-only array

[RhysU]

Code Sample, a copy-pastable example if possible

import numpy as np
import pandas as pd
a = np.array(range(3))
b = ['a', 'b']
# Succeeds, as expected
pd.MultiIndex.from_product([a, b]) 
a.setflags(write=False)
# Raises a ValueError with "buffer source array is read-only"
pd.MultiIndex.from_product([a, b]) 

Problem description

Nothing in the MultiIndex.from_product documentation suggests the inputs need to be mutable. Moreover, it is surprising that the arguments are being mutated.

Expected Output

Not raising.

Output of pd.show_versions()

INSTALLED VERSIONS ------------------ commit: None python: 3.5.2.final.0 python-bits: 64 OS: Linux OS-release: 4.1.17-pv-ts1 machine: x86_64 processor: byteorder: little LC_ALL: en_US.UTF-8 LANG: en_US.UTF-8 LOCALE: en_US.UTF-8

pandas: 0.19.2-ts2
nose: 1.3.7
pip: 9.0.1
setuptools: 27.2.0
Cython: None
numpy: 1.11.3
scipy: 0.18.1
statsmodels: None
xarray: None
IPython: 5.1.0
sphinx: 1.5.1
patsy: None
dateutil: 2.6.0
pytz: 2016.10
blosc: None
bottleneck: None
tables: None
numexpr: None
matplotlib: 1.5.3
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml: None
bs4: None
html5lib: None
httplib2: None
apiclient: None
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: 2.9.4
boto: None
pandas_datareader: None

[jreback]

yeah this is actually raising here.

In [4]: a = np.array(range(3))
   ...: a.setflags(write=False)
   ...: 

In [5]: pd.factorize(a)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-5-11e8991352c5> in <module>()
----> 1 pd.factorize(a)

/Users/jreback/pandas/pandas/core/algorithms.py in factorize(values, sort, order, na_sentinel, size_hint)
    351     uniques = vec_klass()
    352     check_nulls = not is_integer_dtype(values)
--> 353     labels = table.get_labels(vals, uniques, 0, na_sentinel, check_nulls)
    354 
    355     labels = _ensure_platform_int(labels)

/Users/jreback/pandas/pandas/src/hashtable_class_helper.pxi in pandas.hashtable.Int64HashTable.get_labels (pandas/hashtable.c:15207)()
    848 
    849     @cython.boundscheck(False)
--> 850     def get_labels(self, int64_t[:] values, Int64Vector uniques,
    851                    Py_ssize_t count_prior, Py_ssize_t na_sentinel,
    852                    bint check_null=True):

/Users/jreback/pandas/pandas/hashtable.cpython-35m-darwin.so in View.MemoryView.memoryview_cwrapper (pandas/hashtable.c:40659)()

/Users/jreback/pandas/pandas/hashtable.cpython-35m-darwin.so in View.MemoryView.memoryview.__cinit__ (pandas/hashtable.c:36894)()

ValueError: buffer source array is read-only
> /Users/jreback/pandas/stringsource(345)View.MemoryView.memoryview.__cinit__ (pandas/hashtable.c:36894)()

you can fix the same way as here, where we call a signature w/o memoryviews (which
https://github.com/pandas-dev/pandas/blob/master/pandas/src/algos_take_helper.pxi.in#L172

is a patch that we have done in the past: #10070

and here's the ultimate cause: https://mail.python.org/pipermail/cython-devel/2013-February/003384.html

I would actually like to have a better method of fixing this.

[jreback]

cc @wesm

[wesm]

Does this still raise if you use ndarray instead of memoryview?

[jreback]

no it works if we use ndarray (in the signature). That's what we did in take, made the signature ndarray, the then dispatch inside ( to 2 different routines). IIRC it does make a perf difference.