Here is my original script that I made this past January which worked fine back then. So looks like now it doesn't work with newer versions of MATPLOTLIB or newer version of pandas.

http://nbviewer.ipython.org/gist/pybokeh/416b6fb2910475ef8e68

Would you strip this down to a SSCCE, please? Preferably a script with minimal dependencies, e.g., no pandas.

I suspect that this is more datetime64 issues.

OK so my title is misleading, it should say plot_date() doesn't work with x values of type pandas.Timestamp as this simplified example shows:

http://nbviewer.ipython.org/gist/pybokeh/92d28fa729bd38be8b92

It is throwing exception:
FormatterWarning: Exception in image/png formatter: Python int too large to convert to C long

That warning is coming out of IPython code, not out of mpl code. Can you reproduce without ipython and without pandas?

attn @takluyver @jreback can you point someone for your projects at this?

I can not re-produce without pandas. The problem appears to be with pandas Timeseries object which is why I changed the title. Whether or not plot_date() should work with pandas.Timeseries, that I can not say. I believe back in January, my original bitcoin chart script did work, but I would have to duplicate the same environment that I had back then to confirm what the exact data that I had passed in.

This works fine:
from datetime import datetime
import matplotlib.pyplot as plt

d1 = datetime(2001,5,1)
d2 = datetime(2001,5,2)
d3 = datetime(2001,5,3)

dates = []
dates.append(d1)
dates.append(d2)
dates.append(d3)

values = [1,2,3]

plt.plot_date(dates, values, 'r')
plt.show()

type(dates[0]) returns:
datetime.datetime

The change might be because of some fixes to applying units conversion (an
internal thing), which might have broken some things. If I understand
pandas correctly though, dates from a pandas object is just simply numpy's
datetime64[ns] object? I could be wrong, though.

On Mon, Oct 27, 2014 at 11:42 AM, pybokeh notifications@github.com wrote:

I can not re-produce without pandas. The problem appears to be with pandas
Timeseries object which is why I changed the title. Whether or not
plot_date() should work with pandas.Timeseries, that I can not say. I
believe back in January, my original bitcoin chart script did work, but I
would have to duplicate the same environment that I had back then to
confirm what the exact data that I had passed in.

This works fine:
from datetime import datetime
import matplotlib.pyplot as plt

d1 = datetime(2001,5,1)
d2 = datetime(2001,5,2)
d3 = datetime(2001,5,3)

dates = []
dates.append(d1)
dates.append(d2)
dates.append(d3)

values = [1,2,3]

plt.plot_date(dates, values, 'r')
plt.show()

type(dates[0]) returns:
datetime.datetime

—
Reply to this email directly or view it on GitHub
#3727 (comment)
.

Is it possible that the issue is due to a upgrade of Pandas version? Pandas have changed some of their objects to no longer subclass numpy ndarray which has previously broken plotting of pandas object. See http://matplotlib.org/faq/usage_faq.html?highlight=pandas#types-of-inputs-to-plotting-functions

pd.Timestamp is a direct sub-class of datetime.datetime

@tacaswell - the warning comes from IPython, but that's because IPython catches an error in a formatter and turns it into a warning, I believe.

OK, at work I created an environment with pandas 0.14.1 and my bitcoin chart script works. Uninstalled pandas 0.14.1 and then installed pandas 0.15.0. Now, it does not work.

EDIT:
Sorry I should have revealed what my environment was for my issue that I opened.

Python 3.4.2,
pandas 0.15.0
matplotlib 1.4.2

So looks like the issue has to do with something that has apparently changed with pandas 0.15.0.

This is a 'regression' in pandas 0.15 (due to the refactor of Index), see pandas-dev/pandas#8614.
Short story: matplotlib now sees the pandas index as an array of datetime64[ns] values (which are actually very large int64s), instead of an array of Timestamps (which are subclass of datetime.datetime, and can be handled by matplotlib) in previous versions of pandas. So the underlying reason is that matplotlib does not handle datetime64 as date values but as ints.

Running units.registry[np.datetime64] = pd.tseries.converter.DatetimeConverter() should solve it for now. Can you try that?

pandas-dev/pandas#8693 should solve this

Actually, wouldn't it make sense to get this into matplotlib? Supporting
datetime64 was a feature that got lost with John's passing (he understood
the units system the best and we talked about it at SciPy2012).

On Wed, Nov 5, 2014 at 9:44 AM, Joris Van den Bossche <
notifications@github.com> wrote:

pandas-dev/pandas#8693 pandas-dev/pandas#8693 should
solve this

—
Reply to this email directly or view it on GitHub
#3727 (comment)
.

yes certainly (the issue for this: #1097)
and that is what we also discussed briefly in pandas-dev/pandas#8614 (comment), but there is someone who just needs to do this (and depending on what we exactly want (basic datetime64 support, or refactoring of date support using datetime64))

The merged PR (pandas-dev/pandas#8693) just solves this specific issue regarding plotting a pandas DatetimeIndex which worked previously but broke in pandas 0.15.0

All gist links are broken. Is there another example that we can use to confirm the fix provided?

This issue can actually just be closed I think. The actual issue reported here is closed by a PR in pandas in version 0.15.1(so only pandas 0.15.0 has this problem), and the larger issue (better datetime support in matplotlib) is covered by #1097

I re-ran my old examples and confirmed that with latest version of mpl and pandas, all works ok. Closing this issue.

It seems like this problem happens in pandas 0.21.0.

code to reproduce this error

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from matplotlib import dates

t = np.arange(10)
ts = pd.datetime(2017, 11, 16) + pd.to_timedelta(t, 's')
df = pd.DataFrame(t, index=ts)

fig, ax = plt.subplots()
ax.plot(df)

# format time axis
hfmt = dates.DateFormatter('%m/%d %H:%M')
ax.xaxis.set_major_formatter(hfmt)
fig.autofmt_xdate()
plt.show()

The trace

Traceback (most recent call last):

  File "D:\Anaconda3\lib\site-packages\IPython\core\formatters.py", line 332, in __call__
    return printer(obj)

  File "D:\Anaconda3\lib\site-packages\IPython\core\pylabtools.py", line 237, in <lambda>
    png_formatter.for_type(Figure, lambda fig: print_figure(fig, 'png', **kwargs))

  File "D:\Anaconda3\lib\site-packages\IPython\core\pylabtools.py", line 121, in print_figure
    fig.canvas.print_figure(bytes_io, **kw)

  File "D:\Anaconda3\lib\site-packages\matplotlib\backend_bases.py", line 2208, in print_figure
    **kwargs)

  File "D:\Anaconda3\lib\site-packages\matplotlib\backends\backend_agg.py", line 507, in print_png
    FigureCanvasAgg.draw(self)

  File "D:\Anaconda3\lib\site-packages\matplotlib\backends\backend_agg.py", line 430, in draw
    self.figure.draw(self.renderer)

  File "D:\Anaconda3\lib\site-packages\matplotlib\artist.py", line 55, in draw_wrapper
    return draw(artist, renderer, *args, **kwargs)

  File "D:\Anaconda3\lib\site-packages\matplotlib\figure.py", line 1295, in draw
    renderer, self, artists, self.suppressComposite)

  File "D:\Anaconda3\lib\site-packages\matplotlib\image.py", line 138, in _draw_list_compositing_images
    a.draw(renderer)

  File "D:\Anaconda3\lib\site-packages\matplotlib\artist.py", line 55, in draw_wrapper
    return draw(artist, renderer, *args, **kwargs)

  File "D:\Anaconda3\lib\site-packages\matplotlib\axes\_base.py", line 2399, in draw
    mimage._draw_list_compositing_images(renderer, self, artists)

  File "D:\Anaconda3\lib\site-packages\matplotlib\image.py", line 138, in _draw_list_compositing_images
    a.draw(renderer)

  File "D:\Anaconda3\lib\site-packages\matplotlib\artist.py", line 55, in draw_wrapper
    return draw(artist, renderer, *args, **kwargs)

  File "D:\Anaconda3\lib\site-packages\matplotlib\axis.py", line 1133, in draw
    ticks_to_draw = self._update_ticks(renderer)

  File "D:\Anaconda3\lib\site-packages\matplotlib\axis.py", line 974, in _update_ticks
    tick_tups = list(self.iter_ticks())

  File "D:\Anaconda3\lib\site-packages\matplotlib\axis.py", line 921, in iter_ticks
    for i, val in enumerate(majorLocs)]

  File "D:\Anaconda3\lib\site-packages\matplotlib\axis.py", line 921, in <listcomp>
    for i, val in enumerate(majorLocs)]

  File "D:\Anaconda3\lib\site-packages\matplotlib\dates.py", line 538, in __call__
    dt = num2date(x, self.tz)

  File "D:\Anaconda3\lib\site-packages\matplotlib\dates.py", line 441, in num2date
    return _from_ordinalf(x, tz)

  File "D:\Anaconda3\lib\site-packages\matplotlib\dates.py", line 256, in _from_ordinalf
    dt = datetime.datetime.fromordinal(ix).replace(tzinfo=UTC)

OverflowError: Python int too large to convert to C long

<matplotlib.figure.Figure at 0xebecc88>

See #9610 (short story is that pandas no longer automatically registers its unit converters starting from 0.21.0), but we will fix this in upcoming 0.21.1 to at least first give deprecation warning (pandas-dev/pandas#18301)