Geography Reference
In-Depth Information
Best Validation CCR is 70.20% at epoch 146.
Best Validation CCR is 77.53% at epoch 111.
90
90
80
80
70
70
60
60
Train
Validation
Test
50
50
Train
Validation
Test
40
40
0
100
200
300
400
500
0
100
200
300
400
500
Epochs
Epochs
(a) Cartociudad
(b) IDEE-Base
Figure 16.
Correct classification ratios achieved with the neural network for CartoCiudad and
IDEE-Base.
6. Conclusion
Serving pre-generated map image tiles from a server-side cache has become a popular way
of distributing map imagery on the Web. However, in order to achieve an optimal delivery
of online mapping, adequate cache management strategies are needed. These strategies can
benefit of the intrinsic spatial nature of map tiles to improve its performance. During the
startup of the service, or when the cartography is updated, the cache is temporarily empty and
users experiment a poor Quality of Service. In this chapter, a seeding algorithm that populates
the cache based on the history of previous accesses has been proposed. The seeder should
automatically cache tiles until an acceptable QoS is achieved. Then, tiles could be cached
on-demand when they are first requested. This can be improved with short-term prefetching;
anticipating the following tiles that will be requested after a particular request can improve
users' experience. The metatiling approach presented here requests, for a given tile request,
a bigger map image containing adjacent tiles, to the remote WMS backend. Since the user is
expected to pan continuously over the map, those tiles are likely to be requested. Finally, when
the tile cache runs out of space, it is necessary to determine which tiles should be replaced by
the new ones. A cache replacement algorithm based on neural networks has been presented. It
tries to estimate the probability of a tile request occurring before a certain period of time, based
on the following properties of tile requests: recency of reference, frequency of reference, and
size of the referenced tile. Those tiles that are not likely to be requested shortly are considered
as good candidates for replacement.
Acknowledgements
This work has been partially supported by the Spanish Ministry of Science and Innovation
through the project “España Virtual” (ref. CENIT 2008-1030), a FPI research fellowship from
the University of Valladolid (Spain), the National Centre for Geographic Information (CNIG)
and the National Geographic Institute of Spain (IGN).
