### Table V-9: Average Percentage of Extraneous and Missing Edges: E is the average percentage of extraneous edges for the structures learned by an algorithm for a given network. M is the average percentage of missing edges for the structures learned by an algorithm for a given network. These values were calculated as the number of extraneous or missing edges divided by the total number of edges in the true graph. Results were averaged over all sample sizes and all values for k, the number of initial intervals.

2005

### Table 2: Graph G1 = (V;E1). Numerical description of the output of the qp-procedure applied with different values of n and q. See Table 1 for a description of columns.

2006

"... In PAGE 17: ... The boxplots in Figure 8 highlights the great effectiveness of the non-rejection rate in this case. Table2 shows that one can either select the largest graph manageable with standard techniques, choosing in this way a graph with only 12 wrongly removed edges, or select a sparser graph; for instance, the threshold 0.60 gives a graph with 9365 out of 11175 missing edges, absolute error 85 and a 72:94% improvement rate.... ..."

Cited by 3

### Table 3. Extra edge (EE) and missing edge (ME) errors (%) when learning the ALARM network in 10 trials using the TPDA, PC and RAI algorithms

### Table 2: Typical LFDA network: structure penalty, k number of parents, graph:(correct(c)/total(t)/missed(m) edges), RIP violations, best: optimum found;

"... In PAGE 12: ... Therefore the quality of the learned network depends strongly on the selection method. Table2 shows numerical results for problem sizes of n = 49 or n = 50. LFDA uses truncation selection, BOA tournament selection with 50% replacement.... ..."

### Table 1: Alliance codes in the alliance dataset. 0 or NA No alliance

2006

"... In PAGE 4: ...Gibler and Sarkees [2004]). The data are available at cow2.la.psu.edu. For each nations pair, alliance is coded as in Table1 . There are some missing values in the interstate alliance data, and in this study we treat these as missing edges.... In PAGE 4: ... Various methods for imputing the missing values could be considered instead. While the edges are colored by alliance type (see Table1 ), we will consider only the simpli ed graph with binary edges: existence or absence of an alliance. Note that there are missing values (for reasons unknown to us), and we have chosen to encode these as \no alliance quot;.... ..."

### Table 1: Alliance codes in the alliance dataset. 0 or NA No alliance

"... In PAGE 3: ...du/~marchette/igo.tgz. In this latter dataset, there are a total of 214 nations, since while some of these do not have alliances, they do have other attributes such as trade. For each pair of nations, alliance is coded as in Table1 . There are some missing values in the interstate alliance data, and in this study we treat these as missing edges.... In PAGE 3: ... Various methods for imputing the missing values could be considered instead. While the edges are colored by alliance type (see Table1 ), we will consider only the simpli ed graph with binary edges: existence or absence of an alliance. We will construct an alliance graph for each year.... ..."

### Table 5.2 Results on dataset1 of ALARM network (Algorithm B) (M.E. and E.E. stand for missing edges and extra edges respectively. The CI tests are grouped by the cardinalities of their condition-sets.)

1997

Cited by 27

### Table 6 Experimental results of various algorithms on ALARM net data (M.A., E.A. and W.O. stand for missing edges, extra edges and wrongly oriented edges) Algorithm Node Sample Platform Running Results

2001

"... In PAGE 30: ....1.4. Other learning algorithms Many other learning algorithms have attempted to learn this network from a dataset (and sometimes, from a node ordering as well). Table6 below summarizes their performance. (Section 7 summarizes many of these learning systems.... ..."

### Table 5.2 Results on dataset1 of ALARM network (Algorithm B) (M.E. and E.E. stand for missing edges and extra edges respectively. The CI tests are grouped by the cardinalities of their condition-sets.)

### Table 2: The number of missing edges (i.e. false negatives) out of 10 possible, for various reconstruction methods on the Campanulaceae data of Figure 1. MPBE1 through MPBE4 are the four most parsimonious trees by the first phase of the MPBE method. NJ refers to the tree obtained by neighbor- joining on the three distance matrices (these were identical).

2000

Cited by 33