If you can't get stuff done...

... make it look like you meant it by producing a decent plan!

That's pretty much what I'm going to have to do this week. I sized up the task, and I don't think the comparo is going to be complete by Friday. But, I do have a plan! I'll even number the steps to make it look more official.

1) Construct simulated data sets of 10,000 blocks for each of the three distributions I've been featuring in the paper (Normal, Laplace, Cauchy). Assign correlated dimension values to the measures.

2) Run CISS on the simulated data sets.

3) Run the Random sampler on the simulated data sets.

That much might actually happen by Friday.

3) Code the BLB algorithm.

3) Run BLB on the simulated data sets.

4) Observe that all three are pretty much the same for Normal (at least, I hope this is true), but that CISS starts to look better as the tails get heavier.

5) Run the algorithms using query criteria that reference the correlated dimensions, thus raising the variance of the block sums.

6) Note that CISS does way better in this case (at least, I hope this is true).

7) Be really sad if hopes aren't met.

8) Figure out what I did wrong.

9) Fix it.

10) Write up results.