Closed behinger closed 2 years ago
just a quick update: for whatever reason (should have uploaded a manifest... now I know) - This exact instance does not stop any more.
There seems to be other iterations that stop and I will try to extract one and update this.
Ok. found another one..
using JLD2,FileIO,MixedModelsPermutations,MixedModels,Unfold,Random
g = FileIO.load("/store/users/ehinger/unfoldjl_dev/data/bug_threading/bug_threading2.jld2")
H0 = coef(g["mm"])
H0[2] = 0
perm = permutation(MersenneTwister(1),800,g["mm"];β=H0,blup_method=Unfold.olsranefjf,use_threads=false);
Again: Rename it first to "*.jld2" bug_threading2.jld2.zip
Full package list just in case
(unfoldjl_dev) pkg> st
Status `/store/users/ehinger/unfoldjl_dev/Project.toml`
[cbdf2221] AlgebraOfGraphics v0.2.4
[488c2830] BSplines v0.3.2
[20533458] BenchmarkCI v0.1.7
[6e4b80f9] BenchmarkTools v0.5.0
[0a1fb500] BlockDiagonals v0.1.17
[336ed68f] CSV v0.8.4
[13f3f980] CairoMakie v0.3.19
[5ae59095] Colors v0.12.6
[717857b8] DSP v0.6.10
[a93c6f00] DataFrames v0.22.5
[31a5f54b] Debugger v0.6.7
[39dd38d3] Dierckx v0.5.1
[634d3b9d] DrWatson v2.0.1
[5789e2e9] FileIO v1.6.5
[38e38edf] GLM v1.4.1
[e9467ef8] GLMakie v0.1.30
[033835bb] JLD2 v0.4.2
[ee78f7c6] Makie v0.12.0
[ff71e718] MixedModels v3.6.0
[647c4018] MixedModelsPermutations v0.1.2 `dev/MixedModelsPermutations`
[d5ae56c5] MixedModelsSim v0.2.0 `https://github.com/RePsychLing/MixedModelsSim.jl#master`
[b98c9c47] Pipe v1.3.0
[32113eaa] PkgBenchmark v0.2.10
[92933f4c] ProgressMeter v1.5.0
[438e738f] PyCall v1.92.2
[6c5003b2] PyMNE v0.1.2
[6f49c342] RCall v0.13.10
[90137ffa] StaticArrays v1.0.1
[2913bbd2] StatsBase v0.33.4
[3eaba693] StatsModels v0.6.21
[fd094767] Suppressor v0.2.0
[181c99d8] Unfold v0.2.2 `dev/Unfold`
[276b4fcb] WGLMakie v0.3.3
(unfoldjl_dev) pkg>
just tested this at Julia 1.6.1 and same problem (with MixedModelsPermutations 0.1.3) - it does seem to work with 1.7.0-beta2, at least for this iteration. But under 1.7.0.beta-2 I get looooads of crazy console output:
intersect_var at /buildworker/worker/package_linux64/build/src/subtype.c:2219
intersect at /buildworker/worker/package_linux64/build/src/subtype.c:3044
... x 1000
Thus I cannot test for more iterations, too slow :/
MixedModels doesn't work on 1.7 currently -- it's a bug in julia! https://github.com/JuliaLang/julia/issues/40048
I managed to extract a MWE for my bug. Not easy, it only happens every 1000 simulations or so.
(really hope this generalizes to another computer :P) On my computer this runs until
Progress: Progress: 8%|██████████▌ | ETA: 0:00:01
and then goes on foreverImportant: I used the modified olsranef function that puts the scaling to Identity matrix. else it won't work because of singular REs
I noticed that theta is weird:
[0 3*10⁻10 0]
- same weird theta in another run that hangs[0 -8*10⁻6 0]
, but it seems to work for the first N permutations... maybe a hint.Big note: the file is a .jld2 not a .zip => rename due to github file restrictions... bug_threading2.jld2.zip