Time Labels in "Plotting satellite altitude during re-entry" do not match

[prisemut]

In the example plot, the labeling of the time axis is wrong. The 11th of Nov 2013 is not a Thursday. The year is set wrongly, too.
Matplotlib is not converting the time from .toordinal() correctly. When changing it to .utc_datetime() the times are correct.

[brandon-rhodes]

Thanks for letting me know! I should have time to investigate maybe by late next week, and I'll post any updates here as I look for the root problem.

[brandon-rhodes]

Oddly enough the plot here has the correct dates!

https://rhodesmill.org/skyfield/earth-satellites.html#detecting-propagation-errors

So we'll need to look into the difference between how its x axis is plotted versus the plot you've noted has problems:

https://rhodesmill.org/skyfield/example-plots.html#plotting-satellite-altitude-during-re-entry

[prisemut]

The example gets it right by accident. Look at your year, it's 3894 or something like that.

[brandon-rhodes]

I have returned from travel and taken a look — the difference is in matplotlib! It changed internally how it represents dates with numbers. In 3.1.1, they were measured from the year 1:

https://matplotlib.org/3.1.1/api/dates_api.html#matplotlib.dates.date2num

But now are measured by default from 1970:

https://matplotlib.org/3.3.1/api/dates_api.html#matplotlib.dates.date2num

So exactly as you advise, I'll switch to Python datetimes. They would be inefficient for huge amounts of data (since they need an object-per-datapoint), but hopefully folks don't use this example script as their basis if they need big volumes of data.

[brandon-rhodes]

And that almost worked.
When executed as a stand-alone script.
But when run in the doctests, where Pandas is also imported, it broke because of:
matplotlib/matplotlib#12310
pandas-dev/pandas#22859
So I upgraded Pandas.
And now the date is inexplicably back to 3982 Thursday the 11th, even though toordinal() is no longer involved.
I think I’ll now sleep and try again in the morning. 🙂

[brandon-rhodes]

A further Pandas upgrade got things working. Thanks again for pointing out the effect of matplotlib's breaking change; this will now be a stronger example for users to work from!