The problem with the run.BMM

[sergeikharchenko]

Hi again!

Another one bug.

In run.BMM you have fingR::source.norm.distrib function.
If you have for any sediment source zero values of any markers (see example 1), you'll get Inf values after applying log.

Example 1:

  Sample_name Class_decontamination   Mn   Ca    Y   Rb Sulf  As
1          p6                 ist-2 2357 7273 25.0 87.2  966 6.5
2          p7                 ist-2 2401 8785 23.5 83.0 1611 4.4
3          p8                 ist-2  352 7459 27.6 84.7 2359 0.0
4          p9                 ist-2 1239 9026 26.3 87.7 1797 5.1
5         p10                 ist-2 1829 8405 23.5 79.5 2270 5.8

data %>% dplyr::filter(.data[[class]] == lvl) %>% 
        dplyr::select(-setdiff(colnames(data), tracers)) %>% 
        log()

1 7.765145 8.891924 3.218876 4.468204 6.873164 1.871802
2 7.783641 9.080801 3.157000 4.418841 7.384610 1.481605
3 5.863631 8.917177 3.317816 4.439116 7.765993     -Inf
4 7.122060 9.107865 3.269569 4.473922 7.493874 1.629241
5 7.511525 9.036582 3.157000 4.375757 7.727535 1.757858

And then you'll get such covariation matrix:

Mn    0.637151447  0.0248590004 -0.047949531 -0.0044157323 -0.175839871 NaN
Ca    0.024859000  0.0094514885 -0.002541869 -0.0008024629  0.011451322 NaN
Y    -0.047949531 -0.0025418685  0.004970515  0.0017603691  0.005096706 NaN
Rb   -0.004415732 -0.0008024629  0.001760369  0.0016012739 -0.007882461 NaN
Sulf -0.175839871  0.0114513215  0.005096706 -0.0078824606  0.128953348 NaN
As            NaN           NaN          NaN           NaN          NaN NaN

I guess the data.watcher handles with it on the preparation stage, but it doesn't work without uncertainties data. So anyone who try to use the fingR with zero values and without uncertainties (I know, it's not the true way :-)) will face with the error.

Best regards, Sergey Kharchenko

[tchalauxclergue]

Hey!
In deed, such data should have been removed during the data.watcher step, but you're right, it's always better to prevent such issue! I'll try to fix it as soon as possible, along with the other issue you pointed out ;)
Best regard,
Thomas Chalaux-Clergue

[tchalauxclergue]

Hey Sergey, it should now been patched with the latest update (ver. 2.1.2).
The run.BMM function now pre-checks the data for remaining zero values in the sources. The new BMM.gate.keeper function will stop run.BMM if some source samples have zero values, printing a list of the samples and properties involved.
Could you confirm that it is working?

I'll update data.watcher later to take account of this kind of information earlier in the process ;)

Best, Thomas

[sergeikharchenko]

Dear Thomas! Thank you. I'll check the new version as soon as possible after my fieldworks.