Analyzer-like history trees

[tstenner]

Brainvision Analyzer has a nifty feature (especially for aspiring MDs) called 'history tree', where each transformation (and subsequent transformations) are recorded and can be copied to other raw data files.

In implementation details, each transformation is a class with some parameters and outputs (both are stored in the history file). The output data can be either written in whole in the history file (e.g. an FFT) or only some model parameters (filter settings, ICA matrices, …) to calculate the data on-demand.

It's a bit more complicated than the current implementation, but it makes it easier to enforce analysis SOPs while still allowing one-off special case handling.

[cbrnr]

I like the idea. This would require careful thinking though. For example, it could be necessary to group multiple Python statements into one history entry. Furthermore, maybe this kind of graphical history should not replace the current command history, but complement it (because maybe not all commands can be fully mapped to this kind of history).

Also, these trees tend to get very wide. Most of the time, a node has exactly one branch. Is there an alternative model that retains a more linear look?

[tstenner]

If nodes with only one child weren't indented the trees would grow downwards.
Also, if it were possible to define blocks/containers of several steps the boilerplate processing could be represented as a single node.

[cbrnr]

If nodes with only one child weren't indented the trees would grow downwards.

Good idea.

Also, if it were possible to define blocks/containers of several steps the boilerplate processing could be represented as a single node.

True, this would make the whole tree look much cleaner.