Requires pandas, but not documented?

[shirriff]

I'm new to plotly. I installed plotly with pip install plotly==4.5.4 and when I tried to generate a graph, I got 'ImportError: Plotly express requires pandas to be installed.'

The getting-started page says "This package contains everything you need to write figures to standalone HTML files." I looked through the documentation, but couldn't find anywhere that said pandas was required, or how to install pandas.

To reproduce:

>>> import plotly.express as px
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/ken/blog/corona/env/lib/python3.7/site-packages/plotly/express/__init__.py", line 12, in <module>
    Plotly express requires pandas to be installed."""
ImportError: Plotly express requires pandas to be installed.

Maybe this is intentional (see #2125), but I think the documentation should be clearer. Automatically installing pandas would be even better.

[emmanuelle]

Hi @shirriff thank you for reaching out. We want to limit the required dependencies of plotly, and pandas is only an optional dependency which is used in plotly.express only. That said, I totally agree that what happened to you us confusing, and we could

• update the documentation in places to mention this optional dependencies thing (scikit-image is also required for one of the figure factories for example)
• and possibly set up a greater dependency set which would install with pip install "plotly[complete]", like what other packages do (https://docs.dask.org/en/latest/install.html, see https://github.com/dask/dask/blob/master/setup.py for how this is configured).

This could actually be a really great to contribute to plotly for someone interested. (@nicolaskruchten do you concur?)

[h-keisuke]

Hi @emmanuelle

If you say that pandas is a dependency that is only used in plotly.express , wouldn't it be smarter to make plotly.express a separate project ?

Since plotly.express is part of the plotly project, You need pandas for plotly, right?

[nicolaskruchten]

Our reasoning for not making it a hard dependency is that you can still use plotly.io and plotly.graph_objects to make a large number of charts, all without installing pandas or numpy etc.

[pnirtep]

By the way, i've got installed pandas, but i still get this error - http://joxi.ru/EA4DMDzipXPV0A

[neosirap]

By the way, i've got installed pandas, but i still get this error - http://joxi.ru/EA4DMDzipXPV0A

I have the same problem.

[apadman2]

@pnirtep and @neo-sirapat if you import pandas before you import plotly.express it works!

[jason-danforth]

I had the same problem as @pnirtep and @neo-sirapat. I tried creating a new environment and installing pandas first, to no avail. I noticed a link to this URL in my traceback and after installing numPy version 1.19.3 I was able to run Plotly Express.

[nicolaskruchten]

I really don't understand "the problem" that folks are reporting here... If you see the error ImportError: Plotly express requires pandas to be installed. then you need to install pandas. Are folks seeing this error even when pandas is installed?

[jason-danforth]

@nicolaskruchten correct, the problem reported is that we are receiving the error after installing pandas.

[nicolaskruchten]

@jason-danforth thanks for the explanation. This is a pretty different issue from the initial description at the top here but OK, we can work on figuring out what's happening.

Can you confirm that Pandas is fully and correctly installed in the same environment you're trying to access Plotly from when this happens? For example, in the same Python process as the one you're seeing the PX error in, what does this print out?

import pandas as pd
print(pd.__version__)

Can you also tell me about the way you're accessing Python? Via a virtualenv or conda? Which version of Python? Are you in a Jupyter notebook? Via Classic Notebook, JupyterLab or VSCode's extension? What operating system are you on? Etc :)

[jason-danforth]

@nicolaskruchten no problem, though I just want to reiterate that I was able to resolve the issue by reverting to an earlier version of NumPy.

Pandas 1.1.5
I'm running code from the Windows command line in a virtual environment
Python 3.8.3
Operating system:
Windows 10 Pro
Version 20H2
OS build 19042.630

[nicolaskruchten]

Thanks! OK so if reverting numpy solves it then this is pretty unrelated to plotly... Basically it seems like in some cases pandas doesn't get correctly-enough loaded for plotly to use it?

[jason-danforth]

Not exactly. I had no problems using either Pandas or NumPy, but could not run Plotly Express. After attempting to read through the discussion at that URL (and I'll be honest, I understood about 5% of it) I got that there is a larger problem for Windows users related a recent Windows Update and the latest version of Numpy, and that (I suspect) is what is causing the problem for Plotly users. The Microsoft team is anticipating a fix in late January, so you may have users with this problem for some time...

[jslupicki]

@jason-danforth
It seems NumPy with latest RC version (1.20.0rc1) also work on Windows

[TheRajatSethi]

Having the same problem even if pandas already installed.