0014 Interactive views are generated notebooks

Status

Superseded by 0018

Date

2026-09-28

Context

Users check results interactively (zoom into details, step through the images of a stack) all the time. ras_plot / pc_plot do this on pyramids, but need a running python process behind the plot. Writing the plotting code each time is repetitive and a source of mistakes (colours, axes, slider ranges).

A panel / bokeh server was tried: it needs port forwarding, which the VS Code Remote Tunnels connection of the maintainer's cluster does not allow. A notebook kernel runs on the data machine and talks to the editor through the notebook channel, so it works over any VS Code or Jupyter connection.

Decision

moraine view PYRAMID ... -o view.ipynb writes a notebook with one cell per pyramid calling moraine.command.summary.view_pyramid, which chooses from the data:

• colours: the cyclic colorcet colorwheel over (-pi, pi] for phases (complex data, interferograms), viridis over the 1 % - 99 % range of the pyramid statistics otherwise, (0, 1) for booleans; one colour bar shared by the image and points layers;
• axes: range / azimuth with azimuth down for rasters and point clouds on the radar grid (non negative integer bounds), x / y with north up for map coordinates;
• size: the aspect of the scene fitted into 900 x 700 screen pixels, at most 1:4;
• sliders over the images of a stack, with the ranges of the chosen post processing.

quicklook uses the same cyclic colour map.

Consequences

• No server, no port forwarding; agents create the notebook, the user opens it.
• The notebook contains no data, only calls with absolute pyramid paths.

Do not

• Do not write custom plotting code for checking results; improve view_pyramid instead.
• Do not start plot servers for users on remote machines.