plotly on jupyter creates HUGE notebooks file sizes

[urirosenberg]

Hi, we are seeing an issue with plotly v6.0.0 where running in jupyter creates HUGE notebook files, resulting in crashed kernel and inability to save.

Sample code to reproduce the issue (full example can be found here):

import plotly.express as px
import plotly.graph_objects as go
import pandas as pd
import numpy as np

# Create sample data
dates = pd.date_range(start='2024-01-01', end='2024-12-31', freq='M')
values = np.random.randn(12).cumsum()
categories = ['A', 'B', 'C', 'D']
values_bar = [23, 45, 56, 78]

# Line Chart
fig1 = px.line(x=dates, y=values, 
               title='Monthly Trend 2024',
               labels={'x': 'Date', 'y': 'Value'})
fig1.show()

# Scatter Plot with custom styling
fig2 = go.Figure()
fig2.add_trace(go.Scatter(
    x=np.random.randn(100),
    y=np.random.randn(100),
    mode='markers',
    marker=dict(
        size=10,
        color=np.random.randn(100),
        colorscale='Viridis',
        showscale=True
    )
))
fig2.update_layout(title='Scatter Plot with Color Scale')
fig2.show()

# Bar Chart
fig3 = px.bar(x=categories, y=values_bar,
              title='Category Distribution',
              labels={'x': 'Category', 'y': 'Value'},
              color=values_bar,
              color_continuous_scale='Reds')
fig3.show()

Using plotly version 5.24.1, this generates a >1M file.
Using plotly version 6.0.0, this generates a 19M file.

Looking at the v6.0.0 notebook source json file, I see massive amounts of JS code.

[gvwilson]

thanks for the report @urirosenberg - I'll try to get someone to dig into this in the next cycle. (And thanks for providing the notebook - that will help a lot.)

[marthacryan]

Hey @urirosenberg thanks for this! I'm not able to reproduce this, could you send more info about the environment you're running this in?

• Are you running this in JupyterLab, Jupyter notebook, VS Code, or somewhere else? And what version of that are you using?
• Are you running this on a mac? PC? linux?
• What version of python are you using? Are you in a virtual environment? If so which one?

[pfebrer]

I am seeing the same thing in jupyter notebooks with the following environment:

python == 3.9.21
plotly == 6.0.0
notebook == 7.3.2

There are two problems with plotly==6 that I could find:

• When rendering the first plot, plotly==6 includes TWICE the full plotly.js library.
• When rendering other plots, plotly==6 includes the full plotly.js library for each extra plot.

This can be very simply tested in a clean new notebook by running:

import plotly.graph_objects as go

go.Figure()

and then File > Save and Export notebook as > HTML.

If you do this for plotly==6 and plotly==5.24.1, you can then grep for the header of the plotly.js library and you get:

If you render another empty figure in another cell, you get:

This is indeed a huge problem as @urirosenberg says. In our case, we render notebooks with ~40 plotly plots in our documentation site and with plotly==6.0.0 readthedocs was refusing to host it because some html files where 150MB. We had to downgrade to plotly 5.

[pfebrer]

For anyone getting here, a temporary workaround if you don't need to see your notebook offline is to change the default renderer at the top of the notebook:

import plotly.io as pio

pio.renderers.default = "notebook_connected"

[pfebrer]

If it helps with debugging, for the case of plotly==6 with two figures, first it gets loaded as <script type="module"> and then for each figure it gets loaded as <script type="text/javascript">:

grep -B 1 "plotly.js v" plotly6_2emptyfigs.html

[gvwilson]

@marthacryan please have a look when you can.

[marthacryan]

I think the best solution here is to set plotly_mimetype+notebook_connected to the default renderer in notebook environments. The issue here is that we're using the "offline" mode for the renderer. I can open a PR to do that unless there are objections! cc @emilykl @LiamConnors @gvwilson

[gvwilson]

no objection from me - thank you

[pfebrer]

Hi, I can't say whether changing the default to notebook_connected is a good idea or not, but in my opinion if that is the applied change this issue shouldn't be closed.

It would be a bit sweeping the issue under the rug. And the issue itself doesn't look particularly hard to tackle (maybe I'm missing something that makes it more complex than it seems).

[hlvlad]

I agree with @pfebrer, in my case, I need offline mode to be able to send reports via mail. Pre 6.0.0 version produced reports of roughly 5MB, and 6.0.0+ version produces 24MB plots, which is almost 5x regression in size.

[marthacryan]

Sorry forgot to update here: I'm working on adding back a jupyter mime renderer extension to keep those file sizes down while keeping the offline support. PR should be up soon!