Timestamp issues

Community Support
1

I’m running Boken on Ubuntu using python 3.12.3, with the latest pandas and Bokeh

Up until now, I’ve been using timestamps of the format YYYY-MM-DD and that seems to have worked.

However, picking up bokeh after a period of time, I find the following problem. Is this python or panda or Bokeh? Please advise.

I also find now (although I may not have noticed it before) that Pandas is “converting” my timestamps to YYYY-MM-DD HH:MM:SS … although that may be ancillary to the problem

Any debug hints welcome, thanks

File “/home/roger/Projects/TimeLine/env/lib/python3.12/site-packages/bokeh/util/serialization.py”, line 195, in convert_datetime_type
return obj.value / 10**6.0
^^^^^^^^^
File “pandas/_libs/tslibs/timestamps.pyx”, line 339, in pandas._libs.tslibs.timestamps._Timestamp.value.get
OverflowError: Cannot convert Timestamp to nanoseconds without overflow. Use .asm8.view('i8') to cast represent Timestamp in its own unit (here, us).

2

Hi @Roger can you provide a Minimal Reproducible Example? It’s hard to speculate without code to run.

3

It seem like Pandas 3 changed string parsing to default to microsecond resolution and Bokeh will need an update as a result. Without more information from you, is plausible, but not certain that this is the issue you are seeing. If this is the issue, you can try converting to numpy datetimes as a WAR:

if isinstance(obj, pd.Timestamp):
    return convert_datetime_type(obj.to_datetime64())

Edit: here is fix PR for reference Fix serialization of pandas timestamps outside nanosecond range by bryevdv · Pull Request #15472 · bokeh/bokeh · GitHub