search

gherardovarando / gmat-js

js utility to generate graphically constrained matrices
MIT License
0 stars 0 forks source link

Hard-code 0s in the matrix #1

Closed gherardovarando closed 5 years ago

gherardovarando commented 5 years ago

https://github.com/irenecrsn/gmat/issues/18

gherardovarando commented 5 years ago

Done, and also we hard code the 1 in the diagonal.

ghost commented 5 years ago

Maybe because of this the eigenvalue densities are different?

gherardovarando commented 5 years ago

ok yes it could be. Then we should definitely hard code 0s and 1s because matrices gmat-js obtains are less ill-conditioned

ghost commented 5 years ago

OK, but if the results change so much when rounding, then the numeric errors propagated through the algorithm have much more impact than they should. Maybe we should keep this in mind, and compare with what we obtain when orthogonalizing differently.

If for example we are going to do experiments to see the properties of the distribution, these could be importantly biased by those numeric errors (such as how ill conditioned the matrices are).