plotly / plotly/plotly.py

Scattergl (WebGL) lines are not well behaved when zoomed very close

Aperta
#2,881 1 commento 0 reazioni 0 assegnatari Vedi su GitHub

Nessuno ha ancora preso questa issue.

bug P3
Lingua principale
Python
Stelle
18.8k
Fork
2.8k
Merge medio
16h 26m
PR unite (30g)
21

Descrizione

Plotly version 4.11.0
Plotting offline

This was discovered while working on time-series data that covers a large time span (months) but has interesting features at a very low time span (seconds). The original data was defined using pandas dataframes and timestamps but I have removed those references from my example in favor of the python built-in datetime type, though they appear to behave the same.

In the example below, fig1 and fig3 look fine, but the lines and dots in fig2 are not aligned like they should be. The issue is exaggerated with even larger time spans. Although using Scatter() in place of Scattergl() is a workaround for this issue, I would prefer to continue using Scattergl() to take advantage of the increased performance.

Other things I have noticed while investigating:

  • The misalignment is only in the horizontal direction.
  • The lines appear to "jitter" as the figure is scaled on the x-axis (for example, grabbing the right corner of the xaxis using the mouse and dragging it right or left).
  • The issue may be hardware specific. I could not reproduce the issue on my laptop which has an intel CPU with integrated graphics. The issue was discovered on an intel CPU in a desktop using a discrete Radeon GPU.
import datetime
import plotly.graph_objects as go

# Define data:
date_format = '%Y-%m-%d %H:%M:%S'
dates = ['2020-11-01 0:0:0', '2020-11-01 0:0:1', '2020-11-01 0:0:2', '2020-11-01 0:0:3', '2020-11-01 0:0:4', '2020-11-01 0:0:5']
xvals = [datetime.datetime.strptime(this_date, date_format) for this_date in dates]
yvals = [0, 2, 0, 5, 0]

# Add one value on to the start of the data that is far away from the rest of the data
xvals_stretched = [datetime.datetime.strptime('2020-06-01 0:0:0', date_format)] + xvals
yvals_stretched = [0] + yvals

# Set the initial zoom to show the interesting part of the data
x_window_left = datetime.datetime(2020, 10, 31, 23, 59, 55)
x_window_right = datetime.datetime(2020, 11, 1, 0, 0, 10)

# %% A simple plot that works fine with Scattergl
fig1 = go.Figure()
trace = go.Scattergl(x=xvals, y=yvals)
fig1.add_trace(trace)
fig1.update_layout(title=f'fig1: WebGL (time span: {xvals[-1] - xvals[0]})')
fig1.update_xaxes(range=[x_window_left, x_window_right])
fig1.show(renderer='browser')

# %% The same simple plot with one extra datapoint that is far away on the x-axis
fig2 = go.Figure()
trace = go.Scattergl(x=xvals_stretched, y=yvals_stretched)
fig2.add_trace(trace)
fig2.update_layout(title=f'fig2: WebGL (time span: {xvals_stretched[-1] - xvals_stretched[0]})')
fig2.update_xaxes(range=[x_window_left, x_window_right])
fig2.show(renderer='browser')

# %% The stretched plot but without using Scattergl
fig3 = go.Figure()
trace = go.Scatter(x=xvals_stretched, y=yvals_stretched)
fig3.add_trace(trace)
fig3.update_layout(title=f'fig3: SVG (time span: {xvals_stretched[-1] - xvals_stretched[0]})')
fig3.update_xaxes(range=[x_window_left, x_window_right])
fig3.show(renderer='browser')

Running this code should produce three figures illustrating the issue:

plotly scattergl error fig1

plotly scattergl error fig2

plotly scattergl error fig3

Guida per i contributori

Apri la guida per i contributori

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Direzione di ricerca

Esegui il riproduttore Python fornito con le tre figure e confronta i risultati di Scattergl e Scatter mentre esegui lo zoom sull'asse x. Traccia il percorso di rendering di Scattergl per dati datetime con un valore x distante; il lavoro è completato quando linee e marker rimangono allineati orizzontalmente e stabili durante lo zoom e il ridimensionamento, senza perdere i vantaggi prestazionali di Scattergl.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Valutazione

Stack tecnologico
python
Ambito
data-visualization
Tipo di issue
Bug
Difficoltà
4/5
Tempo stimato
3-5 giorni
Stato di attività
Attiva
Chiarezza
Abbastanza chiara
Idoneità per principianti
55/100

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.