Closed sunt05 closed 5 years ago
Original comment by Fredrik Lindberg (Bitbucket: fredrik_ucg, ).
This is a conflict of versions between scipy and numpy in QGIS 2.18.xx. I recommend that you move over to QGIS3 as QGIS 2 will be deprecated this spring.
Original comment by Fredrik Lindberg (Bitbucket: fredrik_ucg, ).
UMEP versions for QGIS2 is no longer being updated.
Original comment by Or Aleksandrowicz (Bitbucket: oraleks, GitHub: oraleks).
Thanks.
I moved to QGIS3, but I am experiencing a very slow performance of UMEP under it. Calculation of SVF for the same DSM took less than 30 min in the QGIS 2.18 version, and more than a day in QGIS3... So meanwhile I stick to QGIS 2.18.
OA
Original comment by Fredrik Lindberg (Bitbucket: fredrik_ucg, ).
That should not be the case. Could you attach your dsm for me to test the performance of the svfcalculator.
Original comment by Or Aleksandrowicz (Bitbucket: oraleks, GitHub: oraleks).
Link to the DSM file: https://technionmail-my.sharepoint.com/:i:/g/personal/oraleks_technion_ac_il/EZmxFcTh_BpAn1PQHK6CQgUBnOOHyj6tFylqYgYIo8DlGQ?e=7XSONx
Tnx,
Or
Original comment by Fredrik Lindberg (Bitbucket: fredrik_ucg, ).
Thanks, What settings are you using in the SVFCalculator for UMEP 3.9? Are you using new option casting 145 shadows or not?
Original comment by Or Aleksandrowicz (Bitbucket: oraleks, GitHub: oraleks).
Hi, on QGIS 3.4 I was using the new option of casting 145 shadows, including the wall aspect and wall height files. On QGIS 2.18 I was using the same wall aspect and wall height files, according to the old procedure. Unless I tick the new option on QGIS 3.4 I cannot include the wall aspect and wall height files in the SVF calculation.
Original comment by Fredrik Lindberg (Bitbucket: fredrik_ucg, ).
Ok, please do not use the new option. That should be as fast(slow) as the QGIS2 version of UMEP. The 145 was a first test and will be updated in the next version of UMEP.
Original comment by Fredrik Lindberg (Bitbucket: fredrik_ucg, ).
You are correct. Thanks for finding this issue. My test show that QGIS3 takes double the time, approximately. I suspect the reason is how PyQt5 make use of threads (i.e. not freezing QGIS while calculating). I See there is a new class called QgsTask that I could try instead. However this will not happen now as I have other more urgent tasks on my agenda.
Original comment by Fredrik Lindberg (Bitbucket: fredrik_ucg, ).
Hi, I found the bit of code that slowed down the SVFCaclulator in QGIS3. In was numpy.maximum() that seem to to much slower in python3 compared to python2. Found a new function numpy.fmax() that increase the code dramatically. In the next update I will update and performance should be the same as QGIS2. The update will be available within a month or so. /Fredrik
Original comment by Fredrik Lindberg (Bitbucket: fredrik_ucg, ).
No, not from my part but if you are interested in contibruting feel free to add GPU capabilities.
Original changes by Fredrik Lindberg (Bitbucket: fredrik_ucg, ).
changed state from "new" to "wontfix"
Original report by Anonymous.
The original report had attachments: UMEP error.JPG
Hi,
I am attempting to install v. 1.8 of UMEP on QGIS 2.18.28 but I get the following error message: cannot import name NUMPY_MKL
I followed the installation instructions catefully, including the "pip install pandas" method - no problems there.
I am using a Windows 10 64bit OS.
V. 1.4 used to work perfectly on the same machine.
Tnx,
Or Aleksandrowicz