[Qgis-user] QNEAT3

[Clemens Raffler]

Hi Magnus,

fist off - thank you for using QNEAT3! I would like to reply to all 4 
points you mentioned:

@Improved feedback: You are right! Unfortunately, I cannot add the 
current progress to the progress bar due to the datastrucutre returned 
by the internal QGIS networkanalysis functions 
(QgsGraphAnalyzer.dijkstra()) but I will try to improve user feedback 
here by reducing the number of messages put out to the log.

@Long calculation time: The speed of QGIS algorithms is highly dependent 
on the size of the input dataset (eg. by the number of vertices (!) that 
are included in the linestrings that make up your graph) . I guess - 
reading your first question - that you input a very large network, 
because calculating the pointcloud seems to take very long in your case. 
Reducing the number of edges in the graph will significantly increase 
the calculation speed (also for multiple points). For enhanced speed you 
may try PostGIS network analysis extension pgrouting which is written 
purely in C and therefore is VERY fast.

@Errors in outputs:  I am aware of the bug you are mentioning, you may 
have a look at the issues section of the QNEAT3 repository 
(https://github.com/root676/QNEAT3/issues/6). I already worked on 
possible solutions but I currently do not have a proper solution for the 
problem. The main blocker here is, that I would  like to avoid 
implementing the distance interpolation in pure Python code (which is 
very slow when working with high resolution rasters). Currently this 
function relies heavily on builtin QGIS C++ classes (namely 
QgsTinInterpolator) which are not perfectly suited for the task... But I 
am open for new ideas on the solution of this bug or code contributions!

@OpenCL support: Currently I don't see any use case where QNEAT3 
performance could benefit from using OpenCL which is - to my knowledge - 
currently only targeted towards rasters (see 
https://github.com/qgis/QGIS-Enhancement-Proposals/issues/121). In 
addition, OpenCL currently seems to be a very experimental codebase in 
QGIS. You may consider sponsoring additional work on OpenCL enhancement.

If any other questions should come up - feel free to contact me!

Cheers,
Clemens