Information Technology Reference
In-Depth Information
the standard deviation of the sonnet's score. They also present the average time for
generating one sonnet,
8
and they are illustrated with the best-scoring sonnet, its score
and domain.
We did the same exercise with a baseline strategy, where there is no evolution.
The best sonnet is not shown because, as expected, it is not very interesting. But we
leave the baseline numbers for comparison purposes (see Table
12.2
).
Comparatively to the other strategies, the baseline generates poems much faster
and uses more diverse patterns, in a trade-off for substantially lower scores. On
diversity, these numbers show that, with the current settings, there is not a pattern
which is constantly used. Each selected pattern is, on average used between 1.7 and
2.2 times, depending on the strategy and the most frequent pattern was never used
more than 20 times in the 100 poems. Ev (Fig.
12.7
) stands out as the strategy using
more distinct patterns, but its average scores and generation time suggest this strategy
is not the best choice for this task. Although G&T+Ev (Fig.
12.9
) generated the best
poemof this exercise, averages show that the plainG&T (Fig.
12.8
) is the strategy that,
more consistently, produces poems with higher scores. Moreover, G&T runs faster.
Fig. 12.6
Sonnet and its contextualization
8
Measured in a MacBook Pro with a Intel i7 CPU and 8GB of RAM, running MacOS X 10.8.
Search WWH ::
Custom Search