4 ms·

Curious if you have evaluated any of the APL family of languages as being useful for your work.

graycat 8y ago
Looked at APL long ago. My guess is that since it is interpretive it would be too slow. I wrote my dissertation code in just quite portable Fortran. As I suggested, I do believe that some progress in execution time could be had from some new machine instructions, new language features to use those instructions, and some compiling help. For the compiling part, the language would have some well defined semantics the compiler could exploit. E.g., for something simple, the semantics could let the compiler exploit the idea of non-inferior sets. E.g., say have 100,000 points in 2-space. Say, just for intuitive visualization, plot them on a standard X-Y coordinate system. Say that the X coordinate is time and the Y direction is fuel. Part of the work is to minimize the cost of time and fuel. At this point in the computation, we don't know how the costs of time and fuel trade off, but we do know that with time held constant, less fuel saves cost, and with fuel held constant, less time saves cost. So, in the plot of the 100,000 options, we look at the lower left, that is, the south-west parts of the 100,000 points. Point (X2,Y2) is inferior to point (X1,Y1) if X1 <= X2 and Y1 <= Y2. That is, point (X2,Y2) is just equal to point (X1,Y1) or is to the upper right, north east of (X1,Y1). If point (X1,Y1) is not inferior to any other of the 100,000 points, then point (X1,Y1) is non-inferior. So, there in the work, can just discard and ignore all the inferior points and work only with the non-inferior points. May have only 100 non-inferior points. Then, presto, bingo, just saved a factor of 1000 in the work. Okay, have sufficiently restricted programming language semantics that the compiler could figure out all that and take advantage of it. When I wrote my code in Fortran, I had to write and call a subroutine to find the non-inferior points -- bummer, that work should be automated in the compiler, but to do that the compiler will need some help from some semantic guarantees. The above is for just two dimensions, but in practice might have a dozen or more. So, what's the fast way with algorithms, programming language semantics, and hardware to find the non-inferior points for a dozen dimensions? Non-inferior points are simple -- lots more is possible.

Looked at APL long ago. My guess is that since it is interpretive it would be too slow. I wrote my dissertation code in just quite portable Fortran. As I suggested, I do believe that some progress in execution time could be had from some new machine instructions, new language features to use those instructions, and some compiling help. For the compiling part, the language would have some well defined semantics the compiler could exploit. E.g., for something simple, the semantics could let the compiler exploit the idea of non-inferior sets. E.g., say have 100,000 points in 2-space. Say, just for intuitive visualization, plot them on a standard X-Y coordinate system. Say that the X coordinate is time and the Y direction is fuel. Part of the work is to minimize the cost of time and fuel. At this point in the computation, we don't know how the costs of time and fuel trade off, but we do know that with time held constant, less fuel saves cost, and with fuel held constant, less time saves cost. So, in the plot of the 100,000 options, we look at the lower left, that is, the south-west parts of the 100,000 points. Point (X2,Y2) is inferior to point (X1,Y1) if X1 <= X2 and Y1 <= Y2. That is, point (X2,Y2) is just equal to point (X1,Y1) or is to the upper right, north east of (X1,Y1). If point (X1,Y1) is not inferior to any other of the 100,000 points, then point (X1,Y1) is non-inferior. So, there in the work, can just discard and ignore all the inferior points and work only with the non-inferior points. May have only 100 non-inferior points. Then, presto, bingo, just saved a factor of 1000 in the work. Okay, have sufficiently restricted programming language semantics that the compiler could figure out all that and take advantage of it. When I wrote my code in Fortran, I had to write and call a subroutine to find the non-inferior points -- bummer, that work should be automated in the compiler, but to do that the compiler will need some help from some semantic guarantees. The above is for just two dimensions, but in practice might have a dozen or more. So, what's the fast way with algorithms, programming language semantics, and hardware to find the non-inferior points for a dozen dimensions? Non-inferior points are simple -- lots more is possible.