NaN values in Volume/Vertex render as black instead of transparent

[mvdoc]

Description

When using Volume.raw or Vertex.raw with manually set alpha channels, any NaN values in the data are rendered as solid black voxels/vertices instead of being transparent. This happens because the RGBA conversion maps NaN to black (R=0, G=0, B=0), and the alpha channel is set independently — so a voxel with NaN data but non-zero alpha shows up as an opaque black region.

This is particularly problematic when plotting thresholded maps where non-significant regions are set to NaN but should appear transparent (showing curvature only).

Minimal example

import cortex
import numpy as np

subject = "fsaverage"  # or any subject in your pycortex db

# Create random data with some NaN values
n_verts = cortex.db.get_surf(subject, "fiducial", "lh")[0].shape[0] * 2
data = np.random.randn(n_verts)
data[::3] = np.nan  # every 3rd vertex is NaN

# Set up alpha: all vertices semi-transparent
alpha = np.ones(n_verts) * 0.8

vtx = cortex.Vertex(data, subject, vmin=-2, vmax=2, cmap="RdBu_r")
vtx = vtx.raw
vtx.alpha = alpha

# NaN vertices render as black instead of transparent
fig = cortex.quickflat.make_figure(vtx, with_curvature=True)
fig.savefig("/tmp/pycortex_nan_black.png", dpi=150)

Expected behavior

NaN values should render as fully transparent (alpha=0), regardless of the alpha channel value. The curvature pattern should show through where data is NaN.

Current workaround

Manually zero out alpha where data is NaN before setting it:

nan_mask = ~np.isfinite(data)
data = np.where(nan_mask, 0, data)
alpha[nan_mask] = 0

Suggestion

The .raw property (or the RGBA conversion) could automatically set alpha=0 for NaN values, so users don't have to handle this manually every time they plot thresholded data.