[BUG] Plotly Express (px) pandas datetime with timezone converted to UTC

[m0wer]

Similar to what happened in #209 Plotly Express converts the Pandas datetime object to UTC instead of respecting its timezone. If the datetime is a string or is converted to pydatetime it works as expected. A reproducible example is available at https://gist.github.com/m0wer/74d5a8901b9499795d13992d29cc2f0c.

A workaround (shown in the example) is to convert the datetime column of the Pandas DataFrame to pydatetime with:

df['date'].dt.to_pydatetime()

[amadeuspzs]

I can confirm this behaviour and also perceive it to be a bug

[nicolaskruchten]

Hmm, it is really unfortunate that Plotly Express behaves differently than Graph Objects here, as px.scatter(df, x='date', y='value') does what is expected. I'll have to think about how big a breaking change this is.

[nicolaskruchten]

Update: px.scatter(df, x='date', y='value') does not do what is expected per the changes we made in the 4.0 release (see https://plotly.com/python/v4-migration/#timezones), but go.Figure(go.Scatter(x=df["date"], y=df["value"])) does in this particular case... digging into why and what we can do about it.

[nicolaskruchten]

I suspect this was introduced in #1934 because pulling the .values out of a series seems to strip the timezone information:

@jonmmease @alexcjohnson @emmanuelle should we fix this? The bug has been in the wild now much longer than the non-buggy code :(

[nicolaskruchten]

Also cc @chriddyp would love your input on this one.

[nicolaskruchten]

I'm thinking we should fix it... the only plots which will break are those where folks made plots with PX on an hourly-or-smaller timescale and fed in data with timezones but really wanted the plot to be in UTC and relied on the fact that it's automatic, which feels like a small set to me.

[alexcjohnson]

I think I agree we should fix it - px and non-px need to behave the same, and if / when we implement real timezone support in plotly.js the fixed version will still be compatible.

[nicolaskruchten]

I should add that this only happens when the input is a pandas datetime column, not a column of ISO strings and not a numpy array of dates or that kind of thing.

[jonmmease]

I would favor fixing this since it is a bug with respect to the v4 change log.

[emmanuelle]

I also think we should fix it.

[nicolaskruchten]

Resolved in 4.10 https://community.plotly.com/t/announcing-plotly-py-4-10-date-axis-improvements-a-faster-px-imshow-and-full-figures-for-development/44685

[amadeuspzs]

Thanks for fixing this I was coming here to file a bug report!