Running in a Docker container - data is still resampled on zoom

[zemmyang]

I have some code that plots PPG data, recorded at a rate of 75Hz over several hours (about 3M+ points). We need to be able to zoom in and see it at a scale of 5s.

We want it as a dash app on a server, so the testing is done on a docker container.

"overview":

zoomed in:

But the exact same code behaves as intended on a jupyter notebook:

Is there some option that's specific to the dash app? I'm not sure what I'm missing here.

Notes that might help:

• specifying default_n_shown_samples in FigureResampler does not seem to help. I get this error (Could not import lttbc; will use a (slower) python alternative.) on the jupyter notebook, but not on docker
• I'm using Scattergl
• relevant packages: dash==2.10.2, plotly_resampler==0.8.3.2, pandas==1.3.5, Flask==2.2.5

---

The Dockerfile:

FROM python:3.11

WORKDIR /app

COPY requirements.txt requirements.txt
RUN pip install -r requirements.txt
RUN pip install dash[diskcache]

COPY . .

RUN chmod u+x ./entrypoint.sh
ENTRYPOINT ["./entrypoint.sh"]

It's running as a Dash app with Flask, entrypoint.sh is just running the Flask stuff and then Flask sets up the Dash app. But running it as a solo Dash app gets me the same behavior

[jonasvdd]

Hi @zemmyang,

This is certainly a relevant issue, thank you for submitting.

• Could you provide a zip with a minimal (standalone) Dash app that uses dummy data (e.g., NumPy random data) for resampling in which the above described behavior occurs?
• Does it work as a dash-app running on your host machine (i.e., without being encapsulated in a docker container)? Also, it may be worthwhile to skim our dash examples

fyi: In ~the next 2 months, we will release rangeslider alike support! 🔥

Kind regards, Jonas

[zemmyang]

Yup, even as a dash app (see below), it's still downsampled.

from dash import Dash, html, dcc
import pandas as pd

import numpy as np

ls = np.linspace(0, 6.28, num=3_000_000, endpoint=True, axis=0)
lss = np.sin(ls*30_000) * np.sin(ls*3)
ppg_df = pd.DataFrame(lss, columns=["PPGValue"])

import plotly.graph_objects as go
from plotly_resampler import FigureResampler
from plotly_resampler.aggregation import MinMaxAggregator

fig = FigureResampler(default_downsampler=MinMaxAggregator(), default_n_shown_samples=1500)
# fig = FigureResampler()

fig.add_trace(
    go.Scattergl(
        y=ppg_df["PPGValue"],
        x=ppg_df.index,
        mode='lines',
        name='PPG',
        hoverinfo='skip'
    ),
    # row=2, col=1
)

app = Dash(__name__)

app.layout = html.Div([
    dcc.Graph(id='graph-content', figure=fig),
])

if __name__ == '__main__':
    app.run_server(debug=True)

But it looks fine on a notebook.

(I'm running this on a windows machine, if that info helps)

[jonasvdd]

@zemmyang,

I can see that you did not take a proper look at our basic dash app example. We use dash callbacks, which leverage the TraceUpdater dash component to efficiently send the resampled data to the server. As in your app, no callbacks are listed at all, this will not resample. I suggest that you take a really close look at this example and aim to understand the inner workings. I think also the documentation should help. (In a notebook, all this fancy functionality happens under the hood)

Additionally, passing your trace data to the hf_x and hf_y argument will speed up the add_trace method significantly.

fyi: also, if you want to support multiple sessions - you should take a look at this example.

Hope this helps you any further, Kind regards, Jonas

[zemmyang]

ugh, sorry - the usage text in the readme made it sound like you just needed to replace Figure with FigureResampler. i feel like the page to the examples could be highlighted a bit more

anyway, it works now. i ran into other possible problems, but i think they're completely separate from this one.

thanks!