"... In PAGE 9: ... There were 6 missing observations. The training strategy used is shown in Table6 . Most respondents attended workshops/tutorials with substantial interaction between instructor and trainee.... ..."

"... In PAGE 19: ... By comparing the accuracy between the CW and PCW testing sets, we test the hypothesis that the causal relationships triggering interstate conflicts have been stable over time. Table2 summarizes the results of the kappa analysis for the cold war period, in the form of significance matrices.26 Those matrices show all the Z values from comparing the kappa coefficients of the different training strategies (off diagonal elements) two at a time, as well as the Z statistic that measures the significance of each individual classification (main diagonal elements).... In PAGE 19: ...ithout it (Rep). Finally, the result of the unbalanced training is shown (Unb). We use the acronyms C, for conflict, and NC, for non-conflict, in the conditional kappa matrices. - - - - - - - - - - - - - - - - - - - - - - - - Table2 about here - - - - - - - - - - - - - - - - - - - - - - - - The importance of constructing a meaningful training set emerges clearly from the Table 2 results. Three classifications, CVRep, Rep, and Unb, are statistically insignificant, since the Z values in the red main diagonal are smaller than 1.... In PAGE 20: ... This is because the duplication of cases in the training sets leads the network to learn too well the training cases, so fitting data noise rather than data structure. And Table2 stresses another important result: CVBal performs statistically better than all the other training sets. Indeed CVBal also achieves better accuracy than Bal (Z= 6.... ..."

"... In PAGE 25: ... According to them, a training strategy orchestrates methods (such as instruction and practice) and tools (such as simulation, feedback, and performance measures) to convey a content. Table2 summarizes the elements of a training strategy based on APT.... ..."

"... In PAGE 3: ... As CTS data is only available at a bandwidth of 4kHz this poses additional questions on the initialisation and training procedure . Table9 shows recognition performance on the icsidev test set using various model training strategies. The baseline CTS systems yield a still reasonable error rate.... ..."

"... In PAGE 8: ... However, as CTS data is only available at a bandwidth of 4kHz this poses additional questions on the initiali- sation and training procedure. Bandwidth and Adaptation Table7 shows recognition performance on the icsidev test set using various model training strategies. The baseline CTS systems yield a still reasonable error rate.... ..."

"... In PAGE 20: ...han Bal (Z= 6.38). This is empirical evidence that the cross-validation prior correction is effective in significantly reducing selection bias in balanced samples. - - - - - - - - - - - - - - - - - - - - - - - - - - - Table3 about here - - - - - - - - - - - - - - - - - - - - - - - - - - - In Table 3, the error matrix of the CVBal technique in the cold war analysis shows that this training strategy achieves the highest individual class accuracy. It correctly predicts 82.... In PAGE 20: ...han Bal (Z= 6.38). This is empirical evidence that the cross-validation prior correction is effective in significantly reducing selection bias in balanced samples. - - - - - - - - - - - - - - - - - - - - - - - - - - - Table 3 about here - - - - - - - - - - - - - - - - - - - - - - - - - - - In Table3 , the error matrix of the CVBal technique in the cold war analysis shows that this training strategy achieves the highest individual class accuracy. It correctly predicts 82.... ..."

"... In PAGE 21: ... To prevent difference in sample size from influencing the result, we performed kappa and conditional kappa analysis together with a pairwise Z test statistic on the CVBal result as a measure of normalized accuracy, thus making the PCW and CW error matrices directly comparable. - - - - - - - - - - - - - - - - - - - - - - - - - - - Tables 4 amp; 5 about here - - - - - - - - - - - - - - - - - - - - - - - - - - - Table5 summarizes the kappa and conditional kappa analysis results in the form of a significance matrix. Whereas the levels of accuracy for the two periods are similar, the overall performance of the CW model is significantly better on the pre-cold war dyads (Z = 2), with the kappa coefficient for the PCW accuracy larger than for the CW (0.... ..."