node13 A SIMD Interpreter for Genetic Programming on Graphics Cards
Mackey-Glass Results
| | | | | | | | | |
| Mean |
|---|
| RMS error |
.036646 |
.036459 |
.036646 |
.036646 |
.036634 |
.036646 |
.036646 |
.036646 |
.036646 |
.036646 |
.036626 |
|---|
| Solution size |
9 |
11 |
9 |
9 |
13 |
9 |
9 |
9 |
9 |
9 |
9.6 |
|---|
| secs |
167.283 |
168.062 |
167.467 |
167.500 |
167.337 |
167.437 |
167.454 |
167.466 |
167.464 |
167.556 |
167.464 |
|---|
- GPU takes on average 3.3 seconds per generation.
- The tight limit on tree size (15) and depth (4) lead to
similar solutions
that are smaller than those reported for tree GP
The GPU GP runs are faster than all but two CPU runs
despite having a population more than
400
times as big and performing full floating point calculations rather
than 8 bit integer ones.