SCALE_NONE No scaling
SCALE_EXTREME Scale to convergence using largest absolute value
SCALE_RANGE Scale based on the simple numerical range
SCALE_MEAN Numerical range-based scaling
SCALE_GEOMETRIC Geometric scaling
SCALE_CURTISREID Curtis-reid scaling
Additionally, the value can be OR-ed with any combination of one of the following values:
SCALE_QUADRATIC
SCALE_LOGARITHMIC Scale to convergence using logarithmic mean of all values
SCALE_USERWEIGHT User can specify scalars SCALE_POWER2 also do Power scaling
SCALE_EQUILIBRATE Make sure that no scaled number is above 1
SCALE_INTEGERS also scaling integer variables
SCALE_DYNUPDATE dynamic update
SCALE_ROWSONLY scale only rows
SCALE_COLSONLY scale only columns
http://web.mit.edu/lpsolve/doc/set_scaling.htm
By 'scaling', I understand that the program will try to bring all the constraint equations to a a common scale. Say we have, x + y <= 2 and 1000a + 2000b <= 200 then we scale and get x + y <=2 and a + b <= 0.2,