Both the previous and current behavior is now possible

For sign_alpha=-1 old initialization fails

+/-x + \beta is not the same for loarithms

Memory bounds were breached in NLOptimization

hard_copy to ensure reparamterized nodes don't affect those already selected

used for reparamterization against residuals

The differences between +/-1.0 should not affect anything now

I missed that one

This is what you actually need

Clp-17.6 had a bug with openBLAS which seems to be fixed now

fixes some issues with /mathscr v /mathcal

For openBLAS it seems like primal feasability is not enough

Those values are now stored

Refactor to Model

See merge request tpurcell/cpp_sisso!25

Update the most used functions

only NameError allowed

python definitions had a misdefinition in it

automatically strip the units from a df header

Clean up the matlab function

Parameterized latex expression and matlab function used the first operator both times in
binary operators

if norm < 1e-10 treat the feature as constant

With parameters the error tolerance should be a bit lower

Hopefully this will stop the slight errors that pop up

Allow not fully optimized features to be accepted
tighten standard for compare features to 5e-9

set use_global to false in all cases

Fixes a minor bug

Should remove any overwrite issues

An option since while it should give a better model, it also prevents
reproducibility unless OMP_NUM_THREADS=1

These tests needed to be fixed for the updated scheme

This should fix some of the issues

Minor changes were needed

Some debug lines removed

Allow for the features to find better minimum

Remove lines used to check if the feature space is correct

These got unstandardized at one point, now it should be back to normal

The feature space is now the same