Last week I got some help with allocating memory for very large IML arrays. The post got deleted by accident so I thought I'd repost to assist those trying to do that.
In short, IML 14.1 supports matrices with up to 231 elements on the Windows operating system.
I upgraded to 14.1 and have sufficient RAM but a simple "y = j(33000,33000,1);" wasn't working for me.
IIRC, Rick Wicklin suggested adding "-MEMSIZE 12G" to the shortcut I start SAS with.
I went a little deeper and edited that configuration into my SASv9.CFG file and now I can allocate large arrays on my Win7 16GB RAM machine. Thanks, Rick!
Incidentally, this lets me run much larger Monty Hall simulations -- up to a quarter of a billion sims -- see attached code & results below based on Rick's "DO Loop" Monty Hall example -- and remember to "switch doors"!
proc iml;
t0 = time();
NumSim = 2.5e8; /* number of simulated games */
car = randfun(NumSim, "Table", {1 1 1}/3); /* unknown door hides car */
guess = j(NumSim, 1, 1); /* WLOG, you guess door=1 */
show = choose(car=2, 3, 2); /* host shows you a goat --> if car behind 2 then 3 else 2 */
switch = choose(show=2, 3, 2); /* the door you could switch to -- if host shows 2 then 3 else 2 */
winIfSwitch = mean(switch=car); /* (P(win | switch) */
winIfStay = mean(guess=car); /* (P(win | do not switch) */
elapsedTime = time() - t0;
print "Monty Hall: NumSim = " NumSim[format=comma12. label=EMPTY] winIfStay[format=8.6 rowname=" p(WinIfStay)" label=EMPTY] winIfSwitch[format=8.6 rowname=" p(WinIfSwitch)" label=EMPTY] " Time in Seconds = " elapsedTime[format=4.1 label=EMPTY];
| Monty Hall: NumSim = 250,000,000 | p(WinIfStay) 0.333347 | p(WinIfSwitch) 0.666653 | Time in Seconds = 11.1 |
