FAIL: test_get_multi2 (pandas.io.tests.test_data.TestGoogle)

[nickray]

pan = web.get_data_google(['GE', 'MSFT', 'INTC'], 'JAN-01-12', 'JAN-31-12')
result = pan.Open.ix['Jan-15-12':'Jan-20-12']

has shape (5, 3), not (4, 3) as claimed in the test:

               GE   INTC   MSFT
Date                          
2012-01-16  17.65    NaN    NaN
2012-01-17  17.85  25.18  28.40
2012-01-18  17.55  25.13  28.31
2012-01-19  17.60  25.52  28.16
2012-01-20  17.60  25.87  28.82

[cpcloud]

do you get this failure consistently? this will be fixed in 0.13, these tests need a little tlc

[nickray]

Yes.

In any case, a week has business days, why would shape[0] be 4 and not 5?

[jreback]

1/16 is MLK day (not a trading day)...prob bad test

[nickray]

Yes, but many assets trade on "non-trading days", and Google is obviously returning quotes for GE.
I find it hard to believe that Jan 2012 data is inconsistent... Why is code released with tests that fail?

[cpcloud]

I find it hard to believe that Jan 2012 data is inconsistent

Here's what I get when I run the two lines you've pasted above

In [14]: paste
pan = web.get_data_google(['GE', 'MSFT', 'INTC'], 'JAN-01-12', 'JAN-31-12')
result = pan.Open.ix['Jan-15-12':'Jan-20-12']

## -- End pasted text --

In [15]: result
Out[15]:
               GE   INTC   MSFT
Date
2012-01-17  18.99  25.18  28.40
2012-01-18  18.58  25.13  28.31
2012-01-19  19.03  25.52  28.16
2012-01-20  18.81  25.87  28.82

Call me crazy, but that looks pretty inconsistent to me.

[nickray]

Calls you crazy ;-)

Well I'm at a loss. Their homepage lists no prices for GE on MLK: http://is.gd/tS7gsC
And I (consistently... g) get a quote of 17.65

The problem is that some people package software in a way that breaks when tests do not pass:
https://aur.archlinux.org/packages/python2-pandas/

So flaky tests like these should probably not be included.

I do still wonder where this inconsistency is coming from?

[jreback]

can you post:
python, numpy versions, os?

[nickray]

Python 2.7.5, IPython 0.13.2, numpy 1.7.1, up-to-date Arch Linux.

I was actually running an Intel MKL-enabled version of numpy 1.7.1, but replacing it with the stock distro package doesn't change the query result.

[cpcloud]

@nickray also are you using master?

i have an almost identical setup, i use arch too

you can easily get this info via the ci/print_versions.py script from the pandas dir

[nickray]

The AUR PKGBUILD is fetching http://pypi.python.org/packages/source/p/pandas/pandas-0.12.0.tar.gz, which I assume is the official release?

Or what do you mean with master?

[cpcloud]

@nickray release is fine

[nickray]

It was suggested on https://aur.archlinux.org/packages/python2-pandas/ that this might be a locale issue.

However, running in an empty "env" doesn't help.
Also the pre-built version by continuum.io returns the same shape (5, 3).

[cpcloud]

@nickray What do you mean an empty env? Here's the output of locale for me:

LANG=en_US.UTF-8
LC_CTYPE=en_US.UTF-8
LC_NUMERIC="en_US.UTF-8"
LC_TIME="en_US.UTF-8"
LC_COLLATE="en_US.UTF-8"
LC_MONETARY="en_US.UTF-8"
LC_MESSAGES="en_US.UTF-8"
LC_PAPER="en_US.UTF-8"
LC_NAME="en_US.UTF-8"
LC_ADDRESS="en_US.UTF-8"
LC_TELEPHONE="en_US.UTF-8"
LC_MEASUREMENT="en_US.UTF-8"
LC_IDENTIFICATION="en_US.UTF-8"
LC_ALL=

continuum.io should respect locale settings and if that's the issue then it won't matter which version of pandas you use

what is your locale? i'll try to repro using your locale

[nickray]

My locale is exactly the same. By empty env I mean "env -i /bin/sh" and then starting python from that.

[jreback]

@nickray did you happen to do your call before 9:30am EST?

maybe it has different behavior while us mkts are closed....

(if it does then its weird....but possible)

[nickray]

@jreback: No influence, same shape 5, 3 now that the markets are open.

[aginiewicz]

As a note - I packaged this Arch package and I checked on 2 machines that all tests pass before I uploaded it. When I've set LANG=C all worked, although I've seen few locale related failures, so that's why I suggested checking locale.

Can it be not locale but location dependent? Because it passes for me on same system with same versions of software (unless some optional dependencies can affect this? I have all optional dependencies installed on my machine) - I've checked again right now and it still pass for me.

[jreback]

related to #4242? same test?

[cpcloud]

this was never resolved... so not sure

[jreback]

hmm...wonder if its locale related

[jreback]

@cpcloud you are looking at these? (the locale issue)

[cpcloud]

Yep
On Oct 4, 2013 4:20 PM, "jreback" [email protected] wrote:

@cpcloud https://github.com/cpcloud you are looking at these? (the
locale issue)

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/4357#issuecomment-25727363
.

[cpcloud]

closed by #5073