Three big review tasks

[leeyap]

OCEAN PARAMETER UNCERTAINTY

@leeyap

• Take existing code, add in four new ocean params (the fluxes between ocean boxes) with no variability at all. Sanity check — confirm nothing changes
• Add 10% s.d. to new parameters, rerun with n=100 to 1000, assess changes

NEW FILTERING

@bpbond

• Get GCP data
• Rework filtering step in code for generality
• Test adding new filters

SUPPLEMENTARY INFO

• Check with KD and build out

Working in the "reviews" branch.

[bpbond]

@leeyap we are using the latest version of the Hector v3_dev branch for this work right?

[leeyap]

@bpbond I believe so!

[leeyap]

Problem:

I updated the params and without the sd, the scripts ran just fine. Now, I'm getting this error when running the calculate-relative-importance step:

  system is computationally singular: reciprocal condition number = 1.05925e-17
Error in solve.default(covg[-1, -1], covg[-1, 1]) : 
  system is computationally singular: reciprocal condition number = 1.05925e-17
Error in solve.default(covg[-1, -1], covg[-1, 1]) : 
  system is computationally singular: reciprocal condition number = 1.05925e-17
Error in calc.relimp.default.intern(object = structure(list(destination_fraction = c(0.347506128859821,  : 
  covg must be 
 a positive definite covariance matrix 
 or a data matrix / data frame with linearly independent columns.
Error in calc.relimp.default.intern(object = structure(list(destination_fraction = c(0.347081533092334,  : 
  covg must be 
 a positive definite covariance matrix 
 or a data matrix / data frame with linearly independent columns.

I'm not sure what's going on here, or what changed...

[bpbond]

This makes sense! If the new parameters have no variability, it makes sense that relative importance step blows up, since it's based on how parameter variability influences the output.

For your purposes, I think just confirm that things look sensible up to the relative importance step, call it good, and proceed.

[leeyap]

Oh, sorry, that wasn't clear, should have said: "Now, running with distributions." I'm getting this error with the outputs from the runs with varying ocean parameters!

[bpbond]

Oh, I misread your post above, sorry. Hmm. Hard to guess; I'd suggest (i) confirm visually that the new parameters are being generated with distributions you expect, and (ii) check input data for the r.i. step — are the new parameters where you expect, with sensible values? After that, not sure, I can look at your branch tomorrow if useful.