Kaleido 1.0.0rc0 can't plot datetimes as axis label correctly when renderer=“png”

Environment

Python 3.12
Plotly 6.0.0rc0
Kaleido 1.0.0rc0
vscode 1.96.0

Problem

in kaleido 1.0.0rc0, when set renderer="png", Plotly plot datetime as int as axis label, when set renderer="plot_mimetype", Plotly plot datetime correctly as axis label.
in Kaleido 0.2.1, no matter renderer="png" or renderer="plot_mimetype", Plotly can both plot datetime correctly as axis label.

import plotly.graph_objects as go
import pandas as pd

periods = 10
x = pd.date_range("20241215T00:00:00", periods=periods, freq="h").to_numpy()
y = list(range(periods))
fig = go.Figure()
fig.add_trace(go.Scatter(x=x, y=y))
fig.show(renderer="png")

import plotly.graph_objects as go
import pandas as pd

periods = 10
x = pd.date_range("20241215T00:00:00", periods=periods, freq="h").to_numpy()
y = list(range(periods))
fig = go.Figure()
fig.add_trace(go.Scatter(x=x, y=y))
fig.show(renderer="plotly_mimetype")

Temporarily workaround

my temporarily workaround is casting datetime to str dtype.

import plotly.graph_objects as go
import pandas as pd

periods = 1000
x = pd.date_range("20241215T00:00:00", periods=periods, freq="h")
y = list(range(periods))
fig = go.Figure()
fig.add_trace(
    go.Scatter(
        x=x.astype(str),
        y=y,
    )
)
fig.show(renderer="png")

Hi! Thanks for this, I'm going to mark it active.

I think its solved in an unreleased but already-merged pr (which may come out tomorrow or the day after).

Just came across this as well trying to switch to 1.0.0.

I am seeing this with every format for fig.write for 1.0.0

fig.write("figure.sv",format="svg")

So the issue title might be a bit misleading.

Does anyone have a resultion on this?

Smame here, I encountered this issue when exporting figures with every format in version 1.0.0
fig.write_image("xxx")
The exported file is not rendered correctly, although the figure displays properly within the Jupyter Notebook output.

It looks like the issue happens specifically when the timestamps have nanosecond resolution, which is what pandas uses by default. Changing the resolution to microseconds does seem to fix this, but it's not ideal for pandas' default behavior to cause issues like this.

It looks like the issue happens specifically when the timestamps have nanosecond resolution, which is what pandas uses by default. Changing the resolution to microseconds does seem to fix this, but it's not ideal for pandas' default behavior to cause issues like this.

When running the same script in the local macOS Jupyter environment, the output is normal. However, issues arise when using it on a remote Linux workstation.

It looks like the issue happens specifically when the timestamps have nanosecond resolution, which is what pandas uses by default. Changing the resolution to microseconds does seem to fix this, but it's not ideal for pandas' default behavior to cause issues like this.

Thanks! This helped me solve my plotly timeline that suddenly started from 1970: df["start"] = df["start"].astype('datetime64[us]').
Before that I tried just .dt.date, but it didn't work.

It looks like the issue happens specifically when the timestamps have nanosecond resolution, which is what pandas uses by default. Changing the resolution to microseconds does seem to fix this, but it's not ideal for pandas' default behavior to cause issues like this.

Thanks! This helped me solve my plotly timeline that suddenly started from 1970: df["start"] = df["start"].astype('datetime64[us]'). Before that I tried just .dt.date, but it didn't work.

Great! Then a start-point for an expensive workaround - until this issue is expertly fixed on the kaleido side - is by looping over columns and (multi-)index of the input data and testing for is_datetime64_ns_dtype:

df = ... # some input DataFrame passed later to plotly express 

# columns
for n in list(df.columns):
    if pd.core.dtypes.common.is_datetime64_ns_dtype(df[n]):
        df[n] = df[n].astype("datetime64[us]")
# index
for n in list(df.index.names):
    if pd.core.dtypes.common.is_datetime64_ns_dtype(
        df.index.get_level_values(df.index.names.index(n))
    ):
        df.index = df.index.set_levels(
            df.index.get_level_values(df.index.names.index(n)).astype("datetime64[us]"),
            level=df.index.names.index(n),
        )

... probably does not go well on duplicate names. Needs improvement.

It looks like the issue happens specifically when the timestamps have nanosecond resolution, which is what pandas uses by default. Changing the resolution to microseconds does seem to fix this, but it's not ideal for pandas' default behavior to cause issues like this.

Thanks! This helped me solve my plotly timeline that suddenly started from 1970: df["start"] = df["start"].astype('datetime64[us]'). Before that I tried just .dt.date, but it didn't work.

Great! Then a start-point for an expensive workaround - until this issue is expertly fixed on the kaleido side - is by looping over columns and (multi-)index of the input data and testing for is_datetime64_ns_dtype:

df = ... # some input DataFrame passed later to plotly express

columns

for n in list(df.columns):
if pd.core.dtypes.common.is_datetime64_ns_dtype(df[n]):
df[n] = df[n].astype("datetime64[us]")

index

for n in list(df.index.names):
if pd.core.dtypes.common.is_datetime64_ns_dtype(
df.index.get_level_values(df.index.names.index(n))
):
df.index = df.index.set_levels(
df.index.get_level_values(df.index.names.index(n)).astype("datetime64[us]"),
level=df.index.names.index(n),
)

... probably does not go well on duplicate names. Needs improvement.

As far as re-casting the columns go, you could probably do something like dt = dt.astype({col: "datetime64[us]" for col in dt.select_dtypes("datetime64")}), or if you want to use narwhals, you could do something like df = nw.from_native(df).with_columns(ncs.by_dtype(nw.Datetime("ns")).cast(nw.Datetime("us"))).to_native().

@Honghe

@Honghe Nerd points for go.Figure()

@Honghe

@Honghe @ayjayt Has this specific issue been resolved? I am not able to reproduce with plotly==6.4.0 and kaleido==1.2.0.

This code:

import pandas as pd
import plotly.graph_objs as go

periods = 10
x = pd.date_range("20241215T00:00:00", periods=periods, freq="h").to_numpy()
y = list(range(periods))
fig = go.Figure()
fig.add_trace(go.Scatter(x=x, y=y))
fig.show(renderer="png")

produces this image:

which looks correct.

However, there are still some similar issues with [ns] precision datetimes in Plotly.py (e.g. plotly/plotly.py#5355) which I am investigating.

I just tested it under kaleido==1.2.0 for all png, jpg, svg, and pdf and got the same image for all- equal to yours. Closing.

(this was likely fixed in v1.2.0!)

Navigation Menu

Repository navigation

Kaleido 1.0.0rc0 can't plot datetimes as axis label correctly when renderer=“png” #236

Environment

Problem

Temporarily workaround

columns

index

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Kaleido 1.0.0rc0 can't plot datetimes as axis label correctly when renderer=“png” #236

Description

Environment

Problem

Temporarily workaround

Activity

ayjayt commented on Dec 16, 2024

valankar commented on Apr 21, 2025

rod7760 commented on Jun 27, 2025

chetanpat93 commented on Aug 5, 2025

forest009839 commented on Sep 12, 2025

BlastoiseBlue commented on Sep 23, 2025

forest009839 commented on Sep 29, 2025

dsjoho commented on Oct 1, 2025

z--m-n commented on Oct 16, 2025

BlastoiseBlue commented on Oct 16, 2025

columns

index

boomshankerx commented on Oct 16, 2025

emilykl commented on Nov 10, 2025

ayjayt commented on Nov 11, 2025

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions