-
-
Notifications
You must be signed in to change notification settings - Fork 2.7k
Open
@StageTired
Image
Description
I upgraded my requirements from plotly==5.9.0 to plotly==6.3.0 and observed a regression. Markers on the plot disappeared without any error.
Simply put, my code takes a figure as input and plots a dot at the end of each line. Simple reproducible example:
import numpy as np import pandas as pd import plotly.graph_objs as pgo import plotly.subplots np.random.seed(42) # random data dates = pd.date_range(start="2025-01-01", periods=3, freq="D") data = np.random.randn(3, 1) df = pd.DataFrame(data, index=dates, columns=["value"]) # this is plotted elsewhere fig = plotly.subplots.make_subplots() for column in df.columns: trace = pgo.Scatter( x=df.index, y=df[column], mode="lines", ) fig.add_trace(trace) # take the figure and # render last point of each line using fig.data for i, trace in enumerate(fig.data): input_x = trace.x[-1] input_y = trace.y[-1] print("Input x:", input_x, type(input_x)) print("Input y:", input_y, type(input_y)) fig.add_scatter( x=(input_x,), y=(input_y,), mode="markers+text", ) print(f"Plotted x value:", fig.data[-1].x) fig.show()
If I run this with different plotly versions I get different type for the timestamp:
Image
With plotly<6, the figure looks like this
Image With plotly>=6, it looks like this Image
I am aware that I can convert numpy datetime back to datetime.datetime to fix my problem but why does plotly treat numpy datetime as a unix timestamp? It's not being converted to an actual date and there is no error either.
Simply put I would expect x axis to be a date and not an impossibly large number:
import numpy as np import plotly.subplots fig = plotly.subplots.make_subplots() x = np.datetime64('2025-09-26T01:00:00.000000000') y = 1.23 fig.add_scatter( x=(x,), y=(y,), mode="markers+text", ) print(f"Plotted x value:", fig.data[-1].x) fig.show()
