### Table 4: Performance of BEAM and GARI with Positive Examples Only

"... In PAGE 7: ... 10 samples were generated for each language. Table4 shows the performance of BEAM and GARI (population = 200, generations = 100) on these data sets. The figures shown are averages for the 10 samples, using only a single run of the GA on each sample.... ..."

### Table 2: Classification Rates for RPNI, GIG, BEAM and GARI

"... In PAGE 6: ... Eventually, no more merges are possible, and the solution with the lowest MML is the final result. Table2 shows the classification rates achieved by the four algorithms using the second (larger) data set. It is clear that the two algorithms that minimize the number of states (RPNI and GIG) generally outperform the two that minimize message length (BEAM and GARI) on this data set.... ..."

### Table 3: Performance of BEAM and GARI when Minimizing Number of States

"... In PAGE 6: ... For BEAM search, merges were tried in the same depth-first order used by RPNI, and the first solutions found using this ordering were kept. Table3 shows the results. This time the performance of both algorithms is about equal with that of RPNI and GIG.... ..."

### Table 3.5--Nigeria: Financial budget for gari preparation by alternative cassava production and processing technologies, 1991

2004

### Table 3.5 presents a financial analysis of four combinations of cassava production and gari preparation technology. This financial analysis shows that farmers who plant local varieties and grate manually earn a modest net profit of 42 Naira (about US$2.50) per ton of gari. Farmers who plant local varieties and use mechanized grating earn 478 Naira (about US$28.00) net profit per ton of gari as compared with a net profit of 339 Naira (about US20.00) per ton of gari by farmers using TMS varieties and manual grating. Cassava farmers benefit more from using labor-saving grating technology than planting TMS varieties. TMS varieties are significantly more profitable when grating is mechanized. For example, farmers planting the TMS varieties and using mechanized grating earned a net profit of 776 Naira (about US$46.00) per ton of gari.

2004

### Table 2 Aggregate and disaggregate information, Relative forecast accuracy: Average RMSFE ratios over AR(p) of year-on-year inflation in percentage points . horizon 1 6 12

2005

"... In PAGE 21: ... The main criterion for the comparison of the forecasts employed in this study, as in a large part of the literature on forecasting, is the root mean square forecast error (RMSFE). Table2 and 5 present the comparison of the relative forecast accuracy measured in terms of 13Alternatively, the methods suggested by Forni et al. (2000) and Forni, Hallin, Lippi amp; Reichlin (2001) might be applied.... In PAGE 22: ...3.2 Aggregate and disaggregate information First we compare methods only based on aggregate information as opposed to forecast methods for the aggregate including disaggregate variables in addition (see Table2 , column for direct forecast for each forecast horizon). Within the framework of the general theory of prediction we have shown that including disaggregate variables in the aggregate model does improve predictability of a variable (see section 2).... In PAGE 28: ... In- cluding the respective component(s) in the forecast model might then lower forecast accuracy by increasing estimation uncertainty. This might help explaining that selection pays according to the results in Table2 where the VARagg,sub Gets outperforms all other models one month ahead. Further- more, correlation among disaggregate components included in the models decline, i.... ..."

### Table 3. Performance for Single Problems

"... In PAGE 4: ... To assess the reliability of these differences, each prob- lem was solved 100 times with each heuristic, using random value selection. The results, shown in Table3 , suggest that there is indeed some structural feature of each problem that makes it more amenable to one heuristic than the other. The original difference was, therefore, not simply due to the va- garies of value selection.... ..."

### Table 1: Results for short runs of ILS in benchmark instances. Given is the average percentage excess over the best known solution averaged over 10 instances of each size. The maximal CPU-time for SAOP, ILS, and MD is given in seconds on a Sun Sparc 5 Workstation; the computation time for NEH and NEH+ls are significantly lower.

1998

"... In PAGE 10: ... For SAOP the maximal number of iterations is given as a21a24a23a26a25a28a27 a33 a33a32a35 a35 a18 a4a3a6a5 a1 a1 a8a7 a31 a35 a35 a3a9a5 a8 a6 a2 a4a10 a31 a35 a35 a3 a17 a35 a35 a35a12a36 [21] and therefore we allow for ILS and MD the same computation time as needed by SAOP. The computational results are given in Table1 . As expected, ILS performs significantly better than MD and significantly improves over the solution quality achieved by NEH or NEH with additional local search or SAOP.... In PAGE 12: ... 30000 Local search iterations for Tabu-NS and each of 10 runs for ILS, for Tabu-T the best result of 3 runs with 50000 local search iterations each is given. For GA-RY we give the average solution quality, averages for GA-RY are taken over 30 runs (taken from Table1 in [24]). See the text for more details.... In PAGE 12: ... Given are the best, the average and the worst makespan obtained for each instance. For Tabu-T only the best results over several runs are given in [27]; for GA-RY we use the average solution quality given in Table1 in [24]. We only present results for instances with a1 a1 a31 a35 as the smaller instances were solved to optimality in almost every run.... In PAGE 14: ... 15000 Local search iterations for Tabu-NS and each of 5 runs for ILS, for Tabu-T the best result of 3 runs with 10000 local search iterations each is given. For GA-RY we give the average solution quality, averages for GA-RY are taken over 30 runs (taken from Table1 in [24]). See the text for more details.... ..."

Cited by 19

### Table 1 in [24]). See the text for more details.

1998

"... In PAGE 10: ... For SAOP the maximal number of iterations is given as a21a24a23a26a25a28a27 a33 a33a32a35 a35 a18 a4a3a6a5 a1 a1 a8a7 a31 a35 a35 a3a9a5 a8 a6 a2 a4a10 a31 a35 a35 a3 a17 a35 a35 a35a12a36 [21] and therefore we allow for ILS and MD the same computation time as needed by SAOP. The computational results are given in Table1 . As expected, ILS performs significantly better than MD and significantly improves over the solution quality achieved by NEH or NEH with additional local search or SAOP.... In PAGE 11: ...Table1 : Results for short runs of ILS in benchmark instances. Given is the average percentage excess over the best known solution averaged over 10 instances of each size.... In PAGE 12: ... Given are the best, the average and the worst makespan obtained for each instance. For Tabu-T only the best results over several runs are given in [27]; for GA-RY we use the average solution quality given in Table1 in [24]. We only present results for instances with a1 a1 a31 a35 as the smaller instances were solved to optimality in almost every run.... In PAGE 14: ... 15000 Local search iterations for Tabu-NS and each of 5 runs for ILS, for Tabu-T the best result of 3 runs with 10000 local search iterations each is given. For GA-RY we give the average solution quality, averages for GA-RY are taken over 30 runs (taken from Table1 in [24]). See the text for more details.... ..."

Cited by 19

### Table 1 in [24]). See the text for more details.

1998

"... In PAGE 10: ... For SAOP the maximal number of iterations is given as a21a24a23a26a25a28a27 a33 a33a32a35 a35 a18 a4a3a6a5 a1 a1 a8a7 a31 a35 a35 a3a9a5 a8 a6 a2 a4a10 a31 a35 a35 a3 a17 a35 a35 a35a12a36 [21] and therefore we allow for ILS and MD the same computation time as needed by SAOP. The computational results are given in Table1 . As expected, ILS performs significantly better than MD and significantly improves over the solution quality achieved by NEH or NEH with additional local search or SAOP.... In PAGE 11: ...Table1 : Results for short runs of ILS in benchmark instances. Given is the average percentage excess over the best known solution averaged over 10 instances of each size.... In PAGE 12: ... 30000 Local search iterations for Tabu-NS and each of 10 runs for ILS, for Tabu-T the best result of 3 runs with 50000 local search iterations each is given. For GA-RY we give the average solution quality, averages for GA-RY are taken over 30 runs (taken from Table1 in [24]). See the text for more details.... In PAGE 12: ... Given are the best, the average and the worst makespan obtained for each instance. For Tabu-T only the best results over several runs are given in [27]; for GA-RY we use the average solution quality given in Table1 in [24]. We only present results for instances with a1 a1 a31 a35 as the smaller instances were solved to optimality in almost every run.... ..."

Cited by 19