The rangeslider seems to be broken

[kmdalton]

Bug report

I tried creating a rangeslider under ipyplot.imshow by passing a keyword argument ("r", 1., 99., 99). When I actuate the slider, I get a complicated traceback and nothing happens. Here's a simple example to reproduce. I was trying to create a percentile based contrast slider.

from matplotlib import pyplot as plt
from mpl_interactions import ipyplot as iplt
import numpy as np


img = plt.imread("https://matplotlib.org/3.3.1/_images/stinkbug.png")


fig,ax = plt.subplots()

def f(prange):
    pmin,pmax = prange
    vmin = np.percentile(img, pmin)
    vmax = np.percentile(img, pmax)
    truncated = img.copy()
    truncated = np.maximum(vmin, truncated)
    truncated = np.minimum(vmax, truncated)
    return truncated

controls = iplt.imshow(f, prange=("r", 1., 99., 99))
plt.show()

I get a traceback like this when I move the sliders:

Traceback (most recent call last):
  File "anaconda/envs/whatever/lib/python3.8/site-packages/matplotlib/cbook/__init__.py", line 224, in process
    func(*args, **kwargs)
  File "anaconda/envs/whatever/lib/python3.8/site-packages/mpl_interactions/helpers.py", line 394, in changeify
    update({"new": val})
  File "anaconda/envs/whatever/lib/python3.8/site-packages/mpl_interactions/controller.py", line 182, in slider_updated
    self._slider_updated(change, key, values)
  File "anaconda/envs/whatever/lib/python3.8/site-packages/mpl_interactions/controller.py", line 156, in _slider_updated
    self.params[key] = values[int(change["new"])]
TypeError: only size-1 arrays can be converted to Python scalars

Version Info
I saw this bug on 0.17.0 and 0.17.1

[ianhi]

I think this would have been naturally fixed by #189 but definitely deserves a fix sooner than that will be finished.

---

This seems to be an issue only with matplotlib sliders as they don't use tuples for their .val attribute and instead use numpy arrays - which is my fault because I wrote them :(

This was introduced by #155 which added the int casting for backwards compatibility with more numpy versions.

Fix incoming - thanks for the report!

[ianhi]

FWIW you can achieve the same percentile thresholding while avoiding any .copy operations like so:

fig, ax = plt.subplots()

def vmin(prange):
    return np.percentile(img, prange[0])

def vmax(prange):
    return np.percentile(img, prange[1])

controls = iplt.imshow(img, prange=("r", 1.0, 99.0, 99), vmin=vmin, vmax=vmax)
plt.show()

which I think should be more efficient as all you're doing is modifying the norm of the imshow instead of mucking around with the underlying data.

---

This also makes it clear how you might want to be able to pass vmin_vmax as a single callable:

def vmin_vmax(prange):
    return np.percentile(img, prange)

and then do iplt.imshow(img, prange(1,99,99), vmin_vmax=vmin_vmax)

I opened #196 about this