[BUG] Legend does not display when trace has no real data points

[mhangaard]

Describe the bug 🖍️
When plotting a trace with empty x and y, the legend does not display. Plotly has limited options for customizing the legend, so an empty trace is useful in combination with legendgroup to customize the legend.

Reproducing the bug 🔍

import numpy as np
import pandas as pd
import plotly.graph_objects as go
from plotly_resampler import register_plotly_resampler, unregister_plotly_resampler

# comment out this line to get native plotly behavior
register_plotly_resampler(mode="auto", default_n_shown_samples=100)

n = 10000
x = np.arange(n)
x_time = pd.date_range("2020-01-01", freq="1s", periods=len(x))
noisy_sine = (3 + np.sin(x / 2000) + np.random.randn(n) / 10) * x / (n / 4)

fig = go.Figure()
fig.add_trace(go.Scattergl(x=[None], y=[None], name="yp2", showlegend=True, legendgroup="yp2"))
fig.add_trace(go.Scattergl(y=noisy_sine + 2, name="yp2", showlegend=False, legendgroup="yp2"))
fig.add_trace(go.Scattergl(y=noisy_sine - 3, name="ym1"))
fig

Expected behavior 🔧
Native plotly behavior is that the legend of the empty trace appears (blue "yp2" on this screenshot). I can rely on the legend group to toggle the red trace by clicking the "yp2" legend.

I would probably not be able to get the prefixing and suffixing, on the other trace in same legendgroup, but that's also ok.

Environment information:

• OS: Ubuntu @ WSL2
• Python environment:
  • Python version: 3.11
  • plotly-resampler environment:
    • Jupyter-lab 4.1.0
    • ipywidgets 8.1.1
• plotly-resampler version: 0.9.2

Additional context
I have the same behavior in a Dash app.

[jonasvdd]

Hi @mhangaard,

Thank you for providing such a detailed report. 👏🏼

I believe your issue is caused by the check_nans argument of plotly-resampler. The tsdownsample aggregators were (until recently - see this PR 🚧 ) not able to handle NaN-float values, as such. We remove NaN from the input traces fed to plotly-resampler. However, as .isna can be time-consuming when adding a lot of traces, we introduced the check_nans property, so that users can control this NaN checking operation when they are certain no NaNs occur.

Given that your "empty" trace is not a large trace, NaNs should actually not be removed and your described use case is indeed a bug!
But given the PR in progress, this will not be fixed, as this will not occur any more, given that NaN handling will now be performed at data-aggregation runtime (and this will never occur for small empty traces).

TLDR: workaround

use the check_nans argument in the add_trace method in plotly-resampler=0.9.2, your described issue should be fixed in plotly-resampler>0.9.2 .

Working example ⬇️ :

import numpy as np
import pandas as pd
import plotly.graph_objects as go
from plotly_resampler import register_plotly_resampler, unregister_plotly_resampler

# comment out this line to get native plotly behavior
register_plotly_resampler(mode="auto", default_n_shown_samples=100)

n = 10000
x = np.arange(n)
x_time = pd.date_range("2020-01-01", freq="1s", periods=len(x))
noisy_sine = (3 + np.sin(x / 2000) + np.random.randn(n) / 10) * x / (n / 4)
fig = go.Figure()

fig.add_trace(
    go.Scattergl(x=[None], y=[None], name="yp2", showlegend=True, legendgroup="yp2"),
    # NOTE: I added this argument (this is not that loosely coupled, but it works for plotly-resampler=0.9.2)
    check_nans=False,
)
fig.add_trace(
    go.Scattergl(y=noisy_sine + 2, name="yp2", showlegend=False, legendgroup="yp2")
)
fig.add_trace(go.Scattergl(y=noisy_sine - 3, name="ym1"))
fig

Kind regards,
Jonas

[jonasvdd]

related: #294

[mhangaard]

It can be argued that Plotly should have more options to customize the legend items, instead of having to hack the legend using an empty trace.

[mhangaard]

fig.add_trace with check_nans=False made it work for me. Thank you @jonasvdd!