Our first and second attempts ran into problems because we didn't accurately consider several things, and used parameters that really don't matter at all (although they seemed somewhat important before). Right now, I'm changing the model so that we don't care about compression but instead are only interested in at what point of error does a fully aware system perform better than a semi-aware human. We also did not include a utility decay factor in my previous simulations, although I remember we had talked about that a long time ago; that's why I have always had a timestamp attached to objects. Without this decay making old shows less useful, since a user can never watch as many shows as could be recorded, the cache simply becomes saturated at some point.

So, I'm going to do some runs this weekend and meet with Ketan on Monday to discuss things. I'm think we'll get some good results this time. The submission date has been pushed back to December 31, so we have some time left to do this right.

Post a comment