Check if correlation matrix is consistent #675
Conversation
| np.linalg.cholesky(matrix) | ||
| except np.linalg.LinAlgError: | ||
| print("Warning: Correlation matrix is not positive semidefinite") | ||
| return False |
There was a problem hiding this comment.
the others checks raise, but this returns False. raise here too?
actually, the other conditions here would raise, and the matrix would not be fixed. there seems to be some inconsistencies here. if we can fix it we should, but we should print warnings. if there is no way to fix it, then we should raise on the outside.
we can also consider:
- print on "small errors"
- raise on "big errors" (diagonal is 2.0 => raise, diagonal is 1.000001 => print and fix)
There was a problem hiding this comment.
I have instead opted to just return True or False. I think it's cleaner. What do you think?
tests/fmudesign/test_designmatrix.py
Outdated
| Added sensitivity : sens6 | ||
| Warning: Correlation matrix is not positive semidefinite | ||
| Input correlation matrix: | ||
| [[1. 0.9 0. 0. ] |
There was a problem hiding this comment.
would the prints be prettier if the same number of digits was always used?
that way the columns will align.
There was a problem hiding this comment.
I've fixed the output to look like the following. What do you think?
expected_output = """Added sensitivity : seed
Added sensitivity : faults
Added sensitivity : velmodel
Added sensitivity : contacts
Added sensitivity : multz
Added sensitivity : sens6
Warning: Correlation matrix is not consistent
Requirements:
- Ones on the diagonal
- Positive semi-definite matrix
Input correlation matrix:
[[1.00 0.90 0.00 0.00]
[0.90 1.00 0.90 0.00]
[0.00 0.90 1.00 0.00]
[0.00 0.00 0.00 1.00]]
Adjusted to nearest consistent correlation matrix:
[[1.00 0.74 0.11 0.00]
[0.74 1.00 0.74 0.00]
[0.11 0.74 1.00 0.00]
[0.00 0.00 0.00 1.00]]
Added sensitivity : sens7
Added sensitivity : sens8
Provided number of background values (11) is smaller than number of realisations for sensitivity ('sens7', 'p10_p90') and parameter PARAM13. Will be filled with default values.
Provided number of background values (11) is smaller than number of realisations for sensitivity ('sens7', 'p10_p90') and parameter PARAM14. Will be filled with default values.
Provided number of background values (11) is smaller than number of realisations for sensitivity ('sens7', 'p10_p90') and parameter PARAM15. Will be filled with default values.
Provided number of background values (11) is smaller than number of realisations for sensitivity ('sens7', 'p10_p90') and parameter PARAM16. Will be filled with default values.
A total of 91 realizations were generated
Designmatrix written to generateddesignmatrix.xlsx"""| raise ValueError("Correlation matrix diagonal elements must be 1") | ||
|
|
||
| # Check elements are between -1 and 1 | ||
| if np.any(matrix > 1 + 1e-8) or np.any(matrix < -1 - 1e-8): |
There was a problem hiding this comment.
perhaps these errors should be propagated out to print with a more detailed description on the element level?
element C[row=1, column=3] = 1.001, but every correlation should be in range [0, 1]
There was a problem hiding this comment.
I removed this test as I think the definition of a consisten correlation matrix is that it's positive semi-definite and has ones on the diagonal. This is a different type of check I think. Might still be useful though. What do you think?
Previous versions of fmudesign let the users know that a nearest correlation matrix is being used. Bringing this functionality back in. Set seed in design_input_onebyone to be able to check stdout in test.
dafeda
force-pushed
the
check-is-consistent
branch
from
145738e to
a247015
Compare
Previous versions of fmudesign let the users know that a nearest correlation matrix is being used.
Bringing this functionality back in.
Set seed in design_input_onebyone to be able to check stdout in test.