I switch to routine tasks like checkin preparation, mail, time tracking, picking up conversations, eating, streching, surfing. If it really takes longer (>>15m) I will begin the next independent task.
But generally you should avoid long modify-test cycles. Use a much smaller data set/matrix/grid size to try out your modifications and if they show what they shoud run with the larger set.
I have more experience with accounting and media data but the principle of reducing the data volume or even generating synthetic test data should apply to most numerical or engineering problems too.
One anecdote by Feynman when he was in Los Alamos is about this kidn of batch processing (then with punch cards). One thing I remember from that is that they ran multiple computations in parallel. But then they got partial and early feedback which not every computation may.
This is the public group instrumental rationality diary for October 16-31.
The poll earlier this month seems to be sufficiently in favor of maintaining the current schedule that extra votes are unlikely to change things much, but if you'd really like to register your opinion, you are welcome to do so here.
Thanks to cata for starting the Group Rationality Diary posts, and to commenters for participating.
Immediate past diary: October 1-15
Next diary: November 1-15
Rationality diaries archive