Higher interpolator dimensions than required
DARTS creates interpolators for various operators (reservoir, user-defined properties, well, rate, mass rate) in physics. Since not all dimensions of interpolators are exposed, the default operator size self.n_ops
is used (see image), it represents the size of self.reservoir_operators
. This is bad practice as interpolators have to interpolate vectors of significantly higher dimension than required (28 instead of 1 in the snapshot) which potentially downgrade performance. As more and more operators are appearing, this is becoming more significant.
I believe we can move interpolators into a separate library (to avoid the long compilation time of the main repo) and expose them for all possible dimensions.
@d.voskov, @mbwapperom and @i-l-s-h-a-t , i am curious what do you think?