rasterize with time series data fails for plotly and matplotlib

buckeye17 commented 8 months ago

ALL software version info

(this library, plus any other relevant software, e.g. bokeh, python, notebook, OS, browser, etc)

Windows 10 VSCode 1.85.2 Python 3.9.13

bokeh 3.3.4 Interactive plots and applications in the browser from Python dash 2.14.2 A Python framework for building reactive web-apps. Developed by Plotly. dash-bootstrap-components 1.5.0 Bootstrap themed components for use in Plotly Dash dash-core-components 2.0.0 Core component suite for Dash dash-html-components 2.0.0 Vanilla HTML components for Dash datashader 0.16.0 Data visualization toolchain based on aggregating into a grid holoviews 1.18.1 Stop plotting your data - annotate your data and let it visualize itself. jupyter-bokeh 3.0.7 A Jupyter extension for rendering Bokeh content. jupyter-client 8.6.0 Jupyter protocol implementation and client libraries jupyter-core 5.5.1 Jupyter core package. A base package on which Jupyter projects rely. jupyterlab-widgets 3.0.9 Jupyter interactive widgets for JupyterLab numpy 1.26.2 Fundamental package for array computing in Python pandas 2.1.4 Powerful data structures for data analysis, time series, and statistics plotly 5.18.0 An open-source, interactive data visualization library for Python

Description of expected behavior and the observed behavior

To visualize long time series data with Datashader in Dash web app. Note that in a Jupyter notebook, the visualization works without the rasterize() wrapper. And the Dash web app graph works if the X-axis is changed from datetimes to integers. So it seems that rasterize() does not work with a time series axis sent through to_dash().

Complete, minimal, self-contained example code that reproduces the issue

There is an example here: https://community.plotly.com/t/dash-holoviews-datashader-datetime-axes-y-u-no-work/54241

Stack traceback and/or browser JavaScript console output

Traceback (most recent call last):
  File "c:\Users\613186\git_projects\Sensor-Analytics-Pipeline\ui\app.py", line 172, in update_graphs
    components = to_dash(app, [scatter], reset_button=True)
  File "C:\Users\613186\git_projects\Sensor-Analytics-Pipeline\.venv\lib\site-packages\holoviews\plotting\plotly\dash.py", line 377, in to_dash
    plot = PlotlyRenderer.get_plot(hvobj)
  File "C:\Users\613186\git_projects\Sensor-Analytics-Pipeline\.venv\lib\site-packages\holoviews\plotting\renderer.py", line 240, in get_plot
    plot.update(init_key)
  File "C:\Users\613186\git_projects\Sensor-Analytics-Pipeline\.venv\lib\site-packages\holoviews\plotting\plot.py", line 955, in update
    return self.initialize_plot()
  File "C:\Users\613186\git_projects\Sensor-Analytics-Pipeline\.venv\lib\site-packages\holoviews\plotting\plotly\element.py", line 130, in initialize_plot
    fig = self.generate_plot(self.keys[-1], ranges, is_geo=is_geo)
  File "C:\Users\613186\git_projects\Sensor-Analytics-Pipeline\.venv\lib\site-packages\holoviews\plotting\plotly\element.py", line 185, in generate_plot
    data = self.get_data(element, ranges, style, is_geo=is_geo)
  File "C:\Users\613186\git_projects\Sensor-Analytics-Pipeline\.venv\lib\site-packages\holoviews\plotting\plotly\raster.py", line 50, in get_data
    x0, y0 = l+dx/2., b+dy/2.
numpy.core._exceptions._UFuncBinaryResolutionError: ufunc 'add' cannot use operands with types dtype('<M8[ns]') and dtype('float64')

hoxbro commented 8 months ago

This seems to be a problem without dash, both with matplotlib and plotly backend.

import holoviews as hv
from holoviews.operation.datashader import rasterize
import numpy as np
import pandas as pd

hv.extension("plotly")

X = pd.date_range(start="1/1/2018", end="1/08/2018", periods=100)
Y = np.linspace(1, 100, 100)
Z = np.random.randn(100, 100)

rasterize(hv.QuadMesh((X, Y, Z)))

jbednar commented 8 months ago

Datashader cannot natively accept datetime types, so I believe the HoloViews Bokeh backend converts the data to an integer format before sending it out for rendering. So presumably that same code would need to be applied to the other backends as well, hopefully in a way that can be shared between them.

holoviz / holoviews