Array conversion error?

[jd-foster]

MWE

import Pkg
Pkg.activate(@__DIR__)
Pkg.add("CondaPkg")
Pkg.add("PythonCall")

using CondaPkg
CondaPkg.add("python")
CondaPkg.add("matplotlib")

using PythonCall
plt = pyimport("matplotlib.pyplot")
plt.plot([1, 2, 3, 4], [1, 4, 2, 3])

results in

ERROR: Python: TypeError: expecting 1 indices, got 2
Python stacktrace:
 [1] __getitem__
 @ /Users/xxx/.julia/packages/PythonCall/XgP8G/src/jlwrap/array.jl:293:16
 [2] _plot_args
 @ matplotlib.axes._base ~/julia/Examples/PythonCall/.CondaPkg/env/lib/python3.10/site-packages/matplotlib/axes/_base.py:504
 [3] __call__
 @ matplotlib.axes._base ~/julia/Examples/PythonCall/.CondaPkg/env/lib/python3.10/site-packages/matplotlib/axes/_base.py:312
 [4] plot
 @ matplotlib.axes._axes ~/julia/Examples/PythonCall/.CondaPkg/env/lib/python3.10/site-packages/matplotlib/axes/_axes.py:1632
 [5] plot
 @ matplotlib.pyplot ~/julia/Examples/PythonCall/.CondaPkg/env/lib/python3.10/site-packages/matplotlib/pyplot.py:2769
Stacktrace:
 [1] pythrow()
 @ PythonCall ~/.julia/packages/PythonCall/XgP8G/src/err.jl:94
...

with

julia> versioninfo()
Julia Version 1.7.1
Commit ac5cc99908 (2021-12-22 19:35 UTC)
Platform Info:
 OS: macOS (x86_64-apple-darwin19.5.0)

[cjdoris]

The offending line of Matplotlib is

x = x[:, np.newaxis]

so it's assuming that the input is a numpy array.

Seems like this is a bug (or undocumented feature) of Matplotlib? It should really do

x = np.asarray(x)

on its inputs to make sure. I guess you will have to do this yourself.

[icweaver]

I could be remembering wrong, but I feel like this used to work here, at least for single arguments. Trying it again seems to throw the same error now (on main):

using PythonCall

@py import matplotlib.pyplot as plt

plt.plot(rand(10))

ERROR: Python: TypeError: expecting 1 indices, got 2
Python stacktrace:
 [1] __getitem__
 @ ~/.julia/packages/PythonCall/7vZkL/src/jlwrap/array.jl:309
 [2] _plot_args
 @ matplotlib.axes._base ~/Projects/pythoncall_main/.CondaPkg/env/lib/python3.10/site-packages/matplotlib/axes/_base.py:506
 [3] __call__
 @ matplotlib.axes._base ~/Projects/pythoncall_main/.CondaPkg/env/lib/python3.10/site-packages/matplotlib/axes/_base.py:312
 [4] plot
 @ matplotlib.axes._axes ~/Projects/pythoncall_main/.CondaPkg/env/lib/python3.10/site-packages/matplotlib/axes/_axes.py:1632
 [5] plot
 @ matplotlib.pyplot ~/Projects/pythoncall_main/.CondaPkg/env/lib/python3.10/site-packages/matplotlib/pyplot.py:2769
Stacktrace:
 [1] pythrow()
 @ PythonCall ~/.julia/packages/PythonCall/7vZkL/src/err.jl:94
 [2] errcheck
 @ ~/.julia/packages/PythonCall/7vZkL/src/err.jl:10 [inlined]
 [3] pycallargs
 @ ~/.julia/packages/PythonCall/7vZkL/src/abstract/object.jl:210 [inlined]
 [4] pycall(f::Py, args::Vector{Float64}; kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
 @ PythonCall ~/.julia/packages/PythonCall/7vZkL/src/abstract/object.jl:228
 [5] pycall
 @ ~/.julia/packages/PythonCall/7vZkL/src/abstract/object.jl:218 [inlined]
 [6] #_#13
 @ ~/.julia/packages/PythonCall/7vZkL/src/Py.jl:358 [inlined]
 [7] (::Py)(args::Vector{Float64})
 @ PythonCall ~/.julia/packages/PythonCall/7vZkL/src/Py.jl:358
 [8] top-level scope
 @ REPL[11]:1

I guess we could just wrap everything in pylists?

[cjdoris]

Yeah it used to work.

The issue is that this function has changed.

The old version (in the link) eventually does np.atleast_1d(x) when x is a Julia array, converting it to a numpy array.

Unfortunately a Julia array looks enough like a numpy array that it passes through the current version of this function unchanged, which is a bug because the output really does need to be a numpy array.

Should be an easy fix, I'll make an issue on Matplotlib.

[cjdoris]

For now you can call np.array on any Julia arrays manually, or pin Matplotlib to an older version.

[icweaver]

Ah, gotcha, thanks for following up!

[jd-foster]

Thank you for verifying the issue, I was not entirely sure it was just my setup.

[cjdoris]

I changed something in PythonCall (wrapped Julia arrays now have a to_numpy() method) which is enough to get matplotlib to do the desired conversion, so this works again. It's released.