CiteSeerX — Table search results

Results 21 - 30 of 21,326

Table XI. Break-Even Point (in Seconds): Illustrates the Time Required for the Optimization to Pay Off. If the Unoptimized Program Version Ran Longer than the Break-Even Point, Performing Trace Scheduling First and then Running the Optimized Program Version would Perform Better Overall. See Accompanying Text for an Explanation on how these Values Are Computed. The Compilation Cost C0 Includes the Cost for Applying Standard Optimizations to the Application and Inserting Instrumentation Utilized Later by the Dynamic Trace Scheduler. C1 Includes the Cost for Reading the Collected Path Profiling Data and Reoptimizing the Application Using the Trace Scheduler that is Guided by the Path Profiles

in Continuous Program Optimization: A Case

by Thomas Kistler, Michael Franz

Table 2: Break-even points: Perceptron versus SVMs

in On the Dual Formulation of Regularized Linear Systems With Convex Risks

by Tong Zhang

"... In PAGE 17: ... The running time for our particular implementation is 191 seconds. Table2 includes the comparison (break-even points) of the three algorithms over all ten categorizes and the micro-averaged break-even points. It can be observed that both SVMs are consistently better than the Perceptron algorithm.... ..."

Table 2: Break-even point for different algorithm and parameter settings

by Janyce Wiebe

"... In PAGE 6: ... We experiment with two different values, namely 100 and 160 top-ranked distributionally similar words. Table2 shows the break-even points for the four different settings that were evaluated,9 with results that are almost double compared to the informed baseline. As it turns out, for weaker versions of the algorithm (i.... ..."

Table 3: Break-even summaries for Reuters-22173

in Context-Sensitive Learning Methods for Text Categorization

by W.W. Cohen, Y. Singer

"... In PAGE 11: ... Performance was further summarized by a break-even point|a hypothetical point, obtained by interpolation, at which precision equals recall. Table3 summarizes these \micro-averaged break-even quot; points for sleeping experts, with tri-gram, bi-grams, and uni-grams; Ripper with and without negative tests, Rocchio, a simple decision tree learning system; and a Bayesian classi er, The last two gures are from Lewis and Ringuette [1994]. As an additional point of comparison we also show the results of duplicating the experi- ments conducted by Apte et al.... In PAGE 14: ...rom 0.796 to 0.809. Methodology. We note that Table3 indicates that there is a substantial di erence between the di culty of the categories in the datasets used by Apte at al. and those used by Lewis and Ringuette; for instance, Rocchio apos;s micro-averaged break-even point is 0.... ..."

Table 7: Break-even summaries for Reuters-22173

in Context-Sensitive Learning Methods for Text Categorization

by William W. Cohen, Yoram Singer

"... In PAGE 20: ... Performance was further summarized by a break-even point|a hypothetical point, obtained by interpolation, at which precision equals recall. Table7 summarizes these \micro-averaged break-even quot; points for sleeping-experts, with three word phrases, two word phrases, and single word phrases; Ripper, with and without negative tests; Rocchio; a simple decision tree learning system; and a Bayesian classi er. (Using four word phrases with sleeping-experts provided no additional improvement over three word phrases on these problems.... In PAGE 20: ....3.2 Discussion Previous work. On this corpus, a number of the learning algorithms of Table7 make use of context. The hypotheses of SWAP-1, for instance, are extremely similar to those of Ripper, and decision trees also use the same notion of context as Ripper.... In PAGE 20: ... (In fact, the decision trees used by Lewis and Ringuette can be converted into a ruleset with negative tests.) As shown in Table7 , the learning algorithms that use context are uniformly better than those which 13To be precise, we discarded from the dataset all documents which were tagged with no topics, legal or otherwise. In particular, documents tagged with the topic word \bypass quot; were included.... In PAGE 22: ...rom 0.798 to 0.811. 3.4 The Reuters-21578 collection We note that Table7 indicates that there is a substantial di erence between the di culty of the categories in the datasets used by Apte at al. and those used by Lewis and Ringuette; for instance, the micro-averaged break-even point of Rocchio apos;s algorithm is 0.... ..."

Table 5 shows the average time statistics that we collected for the seven programs in this study. In every case, the time required to execute and validate all tests is small, and the time required to analyze the base program and modi ed version exceeds the time required to run and validate all tests. In this case, the break-even values (the number of tests that must be omitted in order to realize savings) are lower than those calculated for Study 1, because of the lower costs of analysis associated with the reduced-size test suites. Despite this fact, for each program the break-even value exceeds the number of tests in the test suites for the program. In cases such as these, test selection techniques | no matter how successful at reducing the number of tests that must be run | cannot provide savings. Time to Time to Time to Pct. of Break-

in Empirical Studies of a Safe Regression Test Selection Technique

by Gregg Rothermel, Mary Jean Harrold 1998

"... In PAGE 16: ... Table5 : Study 2: execution times and savings (seconds) for each program averaged over its set of versions. The bottom row lists averages over all 132 versions.... In PAGE 20: ... Other results obtained in this study are similar to those obtained in Study 2. Time statistics on test- execution and test-selection costs, shown in Table 7, are nearly identical to those displayed in Table5 . This is not surprising given the equivalence of subjects and test suite sizes across the two studies.... ..."

Cited by 93