Adaptive Affinity Propagation Clustering in MapReduce Environment - Technologies and Applications of Artificial Intelligence - page 215

Fig. 3. Processing steps of the proposed Adaptive Map/Reduce Affinity Propagation

3 Evaluation and Experiments

There are total nine machines employed in our experimental environment. One ma-
chine serves as the master and the other eight machines are slaves. All of these ma-
chines are with CPU: Intel® Core™2 Quad Processor Q6600 (8M Cache, 2.40 GHz,
1066 MHz FSB), RAM: DDR2-800 2G * 2. We chose five datasets from the UC Ir-
vine Machine Learning Repository [10] and the Yale Face Database [11].

3.1 Accuracy Result

The dimension of the Iris dataset is 4. As shown in Table 1, the AMRAP, MRAP and
the Canopy initialed Map/Reduce K-means method produce similar accuracy when
the dimensionality is low.
The dimension of the Wine quality dataset is 11. As shown in Table 2, the result is
that the precision rate and the recall rate decrease significantly. The proposed
AMRAP method produces more stable clustering output than the MRAP and Canopy
initialed Map/Reduce K-means when the data dimensionality increases.