> As a personal anecdote, my first forays into ML are covered with rand() calls, and in retrospect, it's head-smashingly obvious that this was a Bad Idea. It's

> As a personal anecdote, my first forays into ML are covered with rand() calls, and in retrospect, it's head-smashingly obvious that this was a Bad Idea. It's a perfectly fine idea, except that like you said you needed to keep better control over your seeds, and probably also could have benefitted from some kind of test harness. (Not to mention it sounds like your metrics were a bit underpowered)