$Avg(3,6,10,9,1,3,9,7,9,4,7,9,2,1,2,4,10,8,9,7) = 6$
-$Avg(3,6,10,9,1) = 5.8$ $\approx 6$
- -$Sum\left(\frac{k}{N} Samples\right) \cdot \frac{N}{k} \approx Sum(*)$
- -Sampling lets you approximate aggregate values with orders of magnitude less data.
-$Avg(3,6,10,9,1,3,9,7,9,4,7,9,2,1,2,4,10,8,9,7) = 6$
+$Avg(3,6,10,9,1) = 5.8$ $\approx 6$
+ +$Sum\left(\frac{k}{N} Samples\right) \cdot \frac{N}{k} \approx Sum(*)$
+ +Sampling lets you approximate aggregate values with orders of magnitude less data.
+Classical OLA techniques aren't entirely appropriate.
+If you pick $a$, $b$, and $N$ correctly, then the sequence:
+
$K_i = (a\cdot K_{iā1}+b)\;mod\;N$
+will produce $N$ distinct, pseudorandom integers $K_i \in [0, N)$
+To marginalize $p(\{X_i\})$...
+Make Cyclic Sampling into a composable operator
+$G$ | # | $\sum p_{\psi_2}$ |
---|---|---|
1 | 3 | 0.348 |
2 | 4 | 0.288 |
3 | 4 | 0.350 |
$D$ | $G$ | # | $\sum p_{\psi_1}$ |
---|---|---|---|
0 | 1 | 1 | 0.126 |
1 | 1 | 2 | 0.222 |
0 | 2 | 2 | 0.238 |
1 | 2 | 2 | 0.050 |
0 | 3 | 2 | 0.322 |
1 | 3 | 2 | 0.028 |
$G$ | # | $\sum p_{\psi_2}$ |
---|---|---|
1 | 3 | 0.348 |
2 | 4 | 0.288 |
3 | 4 | 0.350 |
$D$ | $G$ | # | $\sum p_{\psi_1}$ |
---|---|---|---|
0 | 1 | 2 | 0.140 |
1 | 1 | 2 | 0.222 |
0 | 2 | 2 | 0.238 |
1 | 2 | 2 | 0.050 |
0 | 3 | 2 | 0.322 |
1 | 3 | 2 | 0.028 |
$G$ | # | $\sum p_{\psi_2}$ |
---|---|---|
1 | 4 | 0.362 |
2 | 4 | 0.288 |
3 | 4 | 0.350 |
$D$ | $G$ | # | $\sum p_{\psi_1}$ |
---|---|---|---|
0 | 1 | 2 | 0.140 |
1 | 1 | 2 | 0.222 |
0 | 2 | 2 | 0.238 |
1 | 2 | 2 | 0.050 |
0 | 3 | 2 | 0.322 |
1 | 3 | 2 | 0.028 |
There's a bit of extra math to compute $\epsilon-\delta$ bounds by adapting Serfling's results. It's in the paper.
+Student: A common benchmark graph.
+VE is binary: It completes, or it doesn't.
+CS gets early results faster, but is overtaken by LJ.
+LJ is only 3-5x slower than VE.
+LJ converges to an exact result before Gibbs gets an approx.
+On some graphs Gibbs is better, but only marginally.
+