MATLAB and Julia
ArrayView uses the Python environment selected by the host application.
MATLAB
Select a Python environment with ArrayView installed and put
matlab/arrayview.m
on the MATLAB path:
pyenv(Version="/path/to/python")
arrayview(A) % one array
arrayview(A, B) % side by side
arrayview(A, B, 'overlay', mask, 'cmap', 'viridis', 'voxel_size', [1 1 3])
[ha, hb] = arrayview(A, B); % one handle per array
hb.update(B2) % replaces only the second pane
It takes the same options as Python's view(): name, overlay, rgb,
mode, dims, index, cmap, vmin, vmax, log, complex,
voxel_size, window and port. dims and index count from 1, as MATLAB
does. E in the viewer copies the current view as an arrayview(...) line.
Without the wrapper, call Python directly: h = py.arrayview.view(A);
Local MATLAB follows the normal display selection and prefers a native window when available. MATLAB started through SSH or VS Code follows that environment's browser or VS Code route.
The array is converted to NumPy before registration. Keep h when explicit
cleanup is needed:
h.close()
Julia
PythonCall:
using PythonCall
arrayview = pyimport("arrayview")
h = arrayview.view(A)
PyCall:
using PyCall
arrayview = pyimport("arrayview")
h = arrayview.view(A)
Everything Python's view() takes works from Julia too: several arrays side
by side, overlays, colour images and the view options.
ha, hb = arrayview.view(A, B; overlay=mask, cmap="viridis", voxel_size=(1, 1, 3))
hb.update(B2) # replaces only the second pane
dims and index count from 0 here, because this is the Python function.
Several arrays in one call must have the same number of dimensions (their
sizes may differ) and cannot be colour images.
ArrayView copies the arrays to temporary NumPy files and starts or reuses a
separate Python server process. This avoids blocking on Julia's Python lock and
lets the viewer survive after the call returns. The returned value is a Python
ViewHandle wrapper with url, sid, port, update(), and close(); a
several-array call returns one per array.
h.close()
IJulia
Use the same PythonCall or PyCall code in an IJulia notebook. ArrayView displays
an inline iframe by default. Inline display is a side effect; the call may return
nothing instead of a handle.
Display routing
| Host | Default display |
|---|---|
| Local desktop | Native window when available, otherwise browser |
| VS Code terminal | VS Code tab |
| VS Code remote or tunnel | VS Code tab through VS Code's private proxy |
| Jupyter or IJulia | Inline iframe |
| Plain SSH | Browser URL through SSH port forwarding |
For plain SSH, follow Remote. VS Code tunnel proxying is handled privately by the extension; no Public entry in the Ports view is required.
Cleanup
In Python, close a handle directly or use a context manager:
handle = arrayview.view(a)
handle.close()
with arrayview.view(a) as handle:
print(handle.url)
close() releases that viewer session. Calling it again is a no-op. If cleanup
fails, it raises an error and can be retried.
Advanced server setup
Start with view(A). Use a persistent server only for shared or multi-hop
setups:
arrayview --serve
arrayview data.npy
Use --relay PORT only when a reverse SSH tunnel exposes ArrayView on a
different remote port. See Remote.