Next: Near Consistency: Theoretical Considerations Up: General Dependency Networks Previous: Probabilistic Decision Trees for

## Probabilistic Inference With Real Data

We have suggested that, because the local distributions learned from data sets of adequate size will be close to the true underlying distribution and hence almost consistent, the procedures for extracting a joint distribution from a dependency network and for answering probabilistic queries should yield fairly accurate results. Nonetheless, there is a concern. It could be that the application of pseudo-Gibbs sampling amplifies the inconsistencies. That is, it could be that small perturbations from the true conditional distributions could lead to large perturbations from the true joint distribution . If this phenomenon did occur, it is likely that predictions of new data rendered by a dependency network would be inaccurate. In this section, we compare the predictions of dependency networks and Bayesian networks on real data sets as a first examination of this concern. We use four datasets: (1) Sewall/Shah, data from Sewall and Shah (1968) regarding the college plans of high-school seniors, (2) Women and Mathematics (WAM), data regarding women's preferences for a career in Mathematics (Fowlkes, Freeny, and Landwehr, 1988), (3) Digits, images of handwritten digits made available by the US Postal Service Office for Advanced Technology (Frey, Hinton, and Dayan, 1995), and (4) Nielsen, data about whether or not users watched five or more minutes of network TV shows aired during a two-week period in 1995 (made available by Nielsen Media Research). In each of these datasets, all variables are finite. For the digits data, we report results for only two (randomly chosen) digits, 2'' and 6''. Additional details about the datasets are given in Table 1. To evaluate the accuracy of dependency networks on these datasets, we randomly partition each dataset into a training set and a test set used to learn models and evaluate them, respectively. We measure the accuracy of each learned model on the test set using the log score:
 (3)

where is the number of variables in . This score can be thought of as the average number of bits needed to encode the observation of a variable in the test set. Note that we measure how well a dependency network predicts an entire case. We could look at predictions of particular conditional probabilities, but because Algorithm 1 uses products of queries to produce a prediction on a full case, we expect comparisons on individual queries to be similar. In our experiments, we determine the probabilities in Equation 3 from a learned dependency network using Algorithm 1. For each pseudo-Gibbs sampler invoked in Algorithm 1, we average 5000 iterations after a 10-iteration burn-in. For each data set, these Gibbs-sampling parameters yield scores with a range of variation of less than 0.1% across 10 runs starting with different (random) initial states. For comparison, we measure the accuracy of two additional model classes: (1) a Bayesian network, and (2) a Bayesian network with no arcs--a baseline model. When learning the non-baseline Bayesian network, we use the algorithm described in Chickering, Heckerman, and Meek (1997) wherein each local distribution consists of a decision tree with binary splits. We use the same parameter and structure priors as used in the learning of dependency networks. We determine from a Bayesian network using the law of total probability--pseudo-Gibbs sampling is not needed. Probability estimates obtained from both Bayesian networks and dependency networks correspond to the a posteriori mean of the (multinomial) parameters. The results are shown in Table 1. The Bayesian networks produce density estimates that are better than those of dependency networks, but only slightly so. In particular, consider the summary score in the second row from the bottom of the table, , which is the geometric mean of averaged over all cases in a dataset. For Digit2, the data set having the worst dependency-network performance, the dependency network assigns a probability to a case, on (geometric) average, that is only 3% lower than that assigned by the Bayesian network.

 Dataset Sewall/Shah WAM Digit2 Digit6 Nielsen Number of variables 5 6 64 64 203 Training cases 9286 790 700 700 1637 Test cases 1032 400 399 400 1637 Score(BN) 1.274 0.907 0.542 0.422 0.188 Score(DN) 1.277 0.911 0.584 0.454 0.189 Score(BL) 1.382 0.930 0.823 0.752 0.231 Score(DN)-Score(BN) 0.002 0.004 0.042 0.033 0.001 Score(BL)-Score(BN) 0.107 0.022 0.281 0.330 0.044 1.00 1.00 1.03 1.02 1.00 0.02 0.16 0.15 0.10 0.02

That dependency networks are (slightly) less accurate than Bayesian networks is not surprising. In each domain, the number of parameters in the Bayesian network are fewer than the number of parameters in the corresponding dependency network. Consequently, the dependency-network estimates should have higher variance. This explanation is consistent with the observation that, roughly, differences in accuracy are larger for the data sets with smaller sample size. Because dependency networks produce joint probabilities via pseudo-Gibbs sampling whereas Bayesian networks produce joint probabilities via multiplication, non-convergence of sampling is another possible explanation for the greater accuracy of Bayesian networks. The small variances of the pseudo-Gibbs-sampler estimates across multiple runs, however, makes this explanation unlikely. Overall, our experiments suggest that Algorithm 1 applied to inconsistent dependency networks learned from data yields accurate joint probabilities. Finally, let us consider issues of computation. Joint estimates produced by dependency networks require far more computation than do those produced by Bayesian networks. For example, on a 600 MHz Pentium III with 128 MB of memory running the Windows 2000 operating system, the determination of for a case in the Nielsen dataset takes on average 2.0 seconds for the dependency network and 0.0006 seconds for the Bayesian network. Consequently, one should use a Bayesian network in those situations where it is known in advance that joint probabilities are needed. Nonetheless, in situations where general probabilistic inference is needed, exact inference in a Bayesian network is often intractable; and practitioners often turn to Gibbs sampling for inference. In such situations, Bayesian networks afford little computational advantage.

Next: Near Consistency: Theoretical Considerations Up: General Dependency Networks Previous: Probabilistic Decision Trees for
Journal of Machine Learning Research, 2000-10-19