search

Circuitscape / Circuitscape.jl

Algorithms from circuit theory to predict connectivity in heterogeneous landscapes
https://circuitscape.org
MIT License
128 stars 35 forks source link

Assertion error: linear solve did not converge in single precision #362

Closed SVMendoza closed 1 year ago

SVMendoza commented 2 years ago

Hello, I am running circuitscape on Julia Version 1.6.4 on a linux server with 40 processors and 250 GB of RAM.

I'm running all-to-one scenarios. using the option parallel (max_parallel = 40) and I have to make 12 runs (I have 12 friction value scenarios that I want to run), to automate it I call the files in a loop in julia. perform tests and all very well, but when I run the complete files it sends a error. I thought it was a problem of memory allocation to the nodes, but it is not the case. the error message is:

ERROR: LoadError: On worker 31: AssertionError: norm(G * v .- curr, Inf) / norm(v, Inf) < 1.0e-5 Stacktrace: [1] solve_linear_system @ ~/.julia/packages/Circuitscape/GMT6U/src/core.jl:616 [2] macro expansion @ ./timing.jl:287 [inlined] [3] multiple_solve @ ~/.julia/packages/Circuitscape/GMT6U/src/raster/advanced.jl:312 [4] multiple_solver @ ~/.julia/packages/Circuitscape/GMT6U/src/raster/advanced.jl:292 [5] advanced_kernel @ ~/.julia/packages/Circuitscape/GMT6U/src/raster/advanced.jl:205 [6] f @ ~/.julia/packages/Circuitscape/GMT6U/src/raster/onetoall.jl:0 [7] #201 @ ~/.julia/packages/Circuitscape/GMT6U/src/raster/onetoall.jl:158 [8] #106 @ /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.6/Distributed/src/process_messages.jl:278 [9] run_work_thunk @ /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.6/Distributed/src/process_messages.jl:63 [10] macro expansion @ /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.6/Distributed/src/process_messages.jl:278 [inlined] [11] #105 @ ./task.jl:411 Stacktrace:

SVMendoza commented 2 years ago

I don't understand, it's like it does all the calculations and when it's cleaning memory it sends the message. It returns the curmap and voltmap (77 maps equal to the number of nodes). and of course after the error message does not continue with the next run. I attach the first two scenarios. and and all the messages printed on the screen Escen.zip hardcopy.0.txt .

ranjanan commented 2 years ago

Which version of CS are you on? Can you make sure you're on the latest (5.11.2). Running using Pkg; Pkg.update() should update your packages.

ranjanan commented 2 years ago

This error stack looks like it is from an older version.

SVMendoza commented 2 years ago

Thanks Ranjan, I'll check and let you know if I can fix it. Sergio

SVMendoza commented 2 years ago

the version is Circuitscape v5.10.1

ranjanan commented 2 years ago

update it and check please

SVMendoza commented 2 years ago

Dear Ranjan,

the problem persists. I already did the update and the version of circuitscape was updated but it sends the same error message.

[2b7a1792] ‘ Circuitscape v5.10.1 à v5.11.2

20220629_error.txt

ranjanan commented 1 year ago

I am closing this issue for now, because this appears to run in double precision. Single precision support is currently experimental.