-
Notifications
You must be signed in to change notification settings - Fork 0
/
NOTES
54 lines (31 loc) · 1.94 KB
/
NOTES
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
2022-09-24:
based on sigma2, mixing is still not great, but does seem to give stable values
now using a final thin of 8 (in combine_runs.R) to get from 2000 to 500 samples, as o.w. would have 16 GB output file
results sent to Charlie are /var/tmp/paleon/composition-eastern-moretaxa/output/PLScomposition_eastern_2073-42_0.2.nc derived from PLScomposition_eastern_0.2-{1..10}_full.nc
2022-09-24:
got new dataset from Charlie a couple weeks ago (2073-42)
ran 10 runs on Savio (savio2) for 150k iterations; took about 11 days, plus a day or two to generate probabilities
Savio fitting:
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Scientific Linux 7.9 (Nitrogen)
Matrix products: default
BLAS/LAPACK: /global/software/sl-7.x86_64/modules/langs/intel/2016.4.072/compilers_and_libraries_2016.4.258/linux/mkl/lib/intel64_lin/libmkl_gf_lp64.so
locale:
[1] C
attached base packages:
[1] parallel stats graphics grDevices utils datasets methods
[8] base
other attached packages:
[1] doParallel_1.0.16 iterators_1.0.13 foreach_1.5.1
[4] Rcpp_1.0.5 RhpcBLASctl_0.21-247.1 ncdf4_1.19
[7] spam_2.9-1 inline_0.3.19
loaded via a namespace (and not attached):
[1] compiler_4.0.3 tools_4.0.3 dotCall64_1.0-1 codetools_0.2-16
[5] grid_4.0.3
SCF prep:
see packrat.lock-sep2022-scf
2022-08-25: ran 4 runs for 150k iterations; took about a week on scf-sm2 nodes.
I left out beech and sugar maple so need to rerun.
mixing of sigma2 not very good for some taxa, in particular cedar, fir, hard_pitc, white birch, yellow birch. Doesn't seem directly correlated with sample size. Does seem associated with higher sigma2 values, so more smooth surfaces, perhaps not surprisingly.
Looks like I should remove 100 out of 300 iterations (50k out of 150k before thinning) and probably do 10 runs.
Runs take about 7 days on high partition with another 2 days for props sample generation.