G. E. P. Box, W. G. Hunter, and J. S. Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. Series in Probability and Mathematical Statistics. Wiley, New York, NY, USA, edition, 1978.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....2 as organized by the skeptic, after repeating 32 additional experiments, one for each equivalence class instance. The choice of 32 is not arbitrary; it is well known that the di#erence between a t distribution and a normal distribution becomes negligible when the number of samples exceeds 30 [29]. We report solutions for the class PC in terms of two cost functions (costIDs) time to solve and number of implications. As we show later, these costs are highly correlated for all solvers we tested. For each cost, we report results in five columns: initV, minV, meanV, maxV, max min. The ....

....number of backtracks, etc. results listed in the stat file currently include: initial value (associated with the reference formula) median value, mean value, standard deviation, minimum value, maximum value, the max min ratio, 95 confidence interval of the mean (based on t statistics [29]) and the coe#cient of variation. The computation of max min ratio always includes not only the minimum and maximum values generated from class instances but also the initial value associated with the reference formula. solverSum, a program that reads, from a designated location in the ....

G. E. P. Box, W. G. Hunter, and J. S. Hunter. Statistics for experimenters: An Introduction to Design, Data Analysis, and Model Building. John Wiley & Sons, 1978.

....use memory. 5. MRS: same as MRB, but uses only semantic features. 6. MRP: same as MRB, but uses only pictorial features. Six first time PicHunter users, naive as to the experimental purposes, participated in this study. They ran the ex periment in a 6 users x 6 versions Latin square design [33]. Each user went through 15 target searches, terminating the search under the target testing paradigm; all searches terminated successfully. The results of these experiments are shown in Table I. The first row has the average number V of 9 image displays visited before convergence to the target; ....

....V D. All 7 versions were run with the same set of 15 target images, which was different from the set of 15 images of the experiments of section V D. 7 users, who were naive as to the purposes of the experiment and had never used PicHunter before, participated in the 7 x 7 Latin square design [33]. The re sults are shown in Table V, which uses the same notation as that of Tables I and II. The same two experienced users who participated in all the previous experiments also ran a subset of the experiments. The user model in the new version of PicHunter (the results of which are shown in ....

G. E. P. Box, W. G. Hunter, and J. S. Hunter, Statistics for Experimenters: An Introduction to Design, Analysis, and Model Building, J. Wiley and Sons, New York, N.Y., 1978.

....randomized to negate any uncontrollable bias that could confound further our estimates of effects. The following section shows the results of the DOE but does not go deep into the computation of the effects. Further information about design of experiments can be found in books by Wu [2] and Box [3]. Test No. A B C D E Y 1 Y 2 Y AVE 1 1 1 1 1 1 y1 1 y1 2 y1 av 2 1 1 1 1 1 y2 1 y2 2 y2 av 3 1 1 1 1 1 y3 1 y3 2 y3 av 4 1 1 1 1 1 y4 1 y4 2 y4 av 5 1 1 1 1 1 y5 1 y5 2 y5 av 6 1 1 1 1 1 y6 1 y6 2 y6 av 7 1 1 1 1 1 y7 1 y7 2 y7 av 8 1 1 1 1 1 y8 1 y8 2 y8 av 9 1 1 1 1 1 y9 1 y9 2 y9 av ....

Box, G., Hunter, W., Hunter, J., Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building, John Wiley & Sons, Inc., 1978.

....to zero. Techniques from design of experiments can assist us to achieve this goal. 3. 3 Design of experiments The theory of Design Of Experiments (DOE) provides a mathematical basis to select an optimal sample set that allows an uncorrelated estimation of the fit parameters, given a fit template [30]. The number of sampling schemes described in literature is vast: starting from full and fractional factorial design, over Placket Burman and Taguchi schemes, to Latin hypercube and even random design. A sampling scheme that allows an uncorrelated estimation of the coefficients of (11) are ....

G. E. P. Box, W. G. Hunter, and J. S. Hunter, Statistics for experimenters: an introduction to design, analysis and model building.New York: John Wiley & Sons, 1978.

....is identical (see Fig. 1) we will briefly treat it in advance. Fig. 3 illustrates how the set of performance samples is composed. The input samples X k are generated using techniques from Design of Experiments in order to optimize the effectiveness of the sample set in the fitting process [7]. This input vector controls the values of (operating point) device currents and voltages. The Operating Point Driven Solver transforms these values into the corresponding transistor geometries and bias currents and voltages. Together with the analysis cards needed for the performance extraction, ....

George E. P. Box, W. G. Hunter, and J. S. Hunter, Statistics for experimenters: an introduction to design, analysis and model building, John Wiley & Sons, New York, 1978.

....results. 2.2 Basic Terminology Before discussing how experiments are designed, it is necessary to introduce some DOE vocabulary. For a system model, there are model parameters that can be varied. These model parameters are called factors. The values assigned to factors are called levels [7]. For example, in a computer system, the factor cache size can be assigned levels of 32 Mbytes or 64 Mbytes. Here, cache size is assigned numerical values and hence is called a numerical (or quantitative) factor. The factor cache replacement policy can be assigned levels of random or ....

....experimenter has excellent knowledge of the system under study. Often, the system is complex and its behavior is difficult to predict. The best guess method may require many experiments before an acceptable result is achieved. It is also difficult to determine if the acceptable result is optimal [7]. Another popular experimentation strategy is the one factor at a time approach, in which one factor is varied while all other factors remain fixed at a nominal level. The nominal levels can be any values that the experimenter wishes to use as a baseline for comparisons. After each factor has ....

[Article contains additional citation context not shown here]

G. E. P. Box, W. G. Hunter, and J. S. Hunter, Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. New York, New York: John Wiley & Sons, 1978.

....necessary to apply segregated population balances as well [13] which bring additional complexity to the system. Accurate models describing the be havior of recombinant microorganisms are required for the organization of both experimental and literature information, for the design of experiments [14, 15] and for the control and op timization of recombinant processes [16] There are two alternative approaches in modeling of the underlying dynamics. One alternative is to directly incorporate all the relevant mea surables in the model, and the other is to construct a simplified model that relates ....

G. E. P. BOX, W. G. HUNTER, and S. HUNTER, Statistics for Experimenters: An Introduction to Design, Data Analysis and Model Building, Wiley (1978).

....and other experimental design projects are open to registered participants from anywhere. Keywords: collaborative computing, Internet, experimen tal de sign, algorithms. 1 INTRODUCTION Experimental design is a well defined discipline in agricul ture, biomedical research, and manufacturing [1]. This is F. Brglez, H. Lavana, D. Ghosh, and R. Kurve have been supported by contracts from the DARPA ARO (P 3316 EL DAAH04 94 G2080 and DAAG55 97 1 0345) and a grant from Semiconductor Re search Corporation. Permission to make digital hard copy of all or part of this work for personal or ....

.... arrangements of the 2 layer graph is 522, according to the formula given in [17] Indeed, evaluating the 64 instances in the class returns a nearly normal distribution of crossing numbers with a sample average of 531.8 and a standard deviation of 45.6, giving rise to 95 confidence interval [1] for the true mean: 520.4, 543.1] The expected crossing number of 522, predicted by the formula, is clearly in this interval. The behavior of the two algorithms, denoted as Treatment 12 (implementing the algorithm dot as described in [18] and Treatment 17 (implementing the algorithm GBFS BC ....

G.E.P. Box, W. G. Hunter, and J. S. Hunter. Statistics for experimenters: An Introduction to Design, Data Analysis, and Model Building. John Wiley Sons, 1978.

....output variable to the levels of a number of predictors, or input variables, that affect it. 16] RSM has been a successful technique for efficiently building and optimizing empirical models of continuous functions since the 1950 s in chemical and mechanical engineering, chemistry, and agriculture [17]. The use of RSM provides significant insight to a previously unknown or complicated response behavior in an efficient manner. RSM approximates the dependency of output metrics to input parameters with an empirical polynomial relationship. In general, the approximation is a second order Taylor ....

Box, G.E.P., Hunter, W.G., Hunter, J.S., Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building, Wiley & Sons, New York, 1978.

....number of implications, number ofbacktracks, etc. results listed in the stat file currently include: initial value (associated with the reference formula) median value, mean value, standard deviation, minimum value, maximum value, 95 confidence interval of the mean (based on the t statistics [32]) the standard coe#cient of variation, and the max min ratio. The computation of max min ratio always includes not only the minimum and maximum values generated from class instances but also the initial value associated with the reference formula. solverSum, a program that reads, from a ....

G. E. P. Box, W. G. Hunter, and J. S. Hunter. Statistics for experimenters: An Introduction to Design, Data Analysis, and Model Building. John Wiley & Sons, 1978.

.... instantaneous signal to noise ratio, # c [m] of each coded bit is # distributed with 2 degrees of freedom, f # c (# c ) 1 e # c # c (12) the equivalent instantaneous signal to noise ratio, #[m] of d combined coded bits will also be # distributed, but with 2d degrees of freedom [3], f # (#) 1 (d 1) # d 1 e # # (13) The pair wise error probability can now be expressed as in equation (3) and is given by P 2 (d) Note that equations (7) 14) are exact expressions. Except for the obvious over bounding of the BER, made by applying the union bound in ....

G.E.P. Box, W.G. Hunter, and J.S. Hunter, Statistics for Experimenters: An Introduction to Design, Data Analysis and Model Building, John Wiley & Sons, 1978.

.... instantaneous signal to noise ratio, # c [m] of each coded bit is # distributed with 2 degrees of freedom, f # c (# c ) 1 c # c (11) the equivalent instantaneous signal to noise ratio, #[m] of d combined coded bits will also be # distributed, but with 2d degrees of freedom [3], f # (#) 1 (d 1) # d 1 e # # (12) The pair wise error probability can now be expressed as in equation (2) and is given by P 2 (d) Note that equations (6) 13) are exact expressions. Except for the obvious over bounding of the BER, made by applying the union bound in ....

G.E.P. Box, W.G. Hunter, and J.S. Hunter, Statistics for Experimenters: An Introduction to Design, Data Analysis and Model Building, John Wiley & Sons, 1978.

....algorithms under test. Statistics are based on the number of instances in each equivalence class and are reported for all common parameters described above: median value, mean value, standard deviation, minimum value, maximum value, 95 confidence interval of the mean (based on the t statistics [5]) and the standard coe#cient of variation. This list can be extended in the future. The report also includes the initial value as a statistics that refers to the parameters evaluated for the reference formula. Ideally, the initial value will always be bounded by the minimum and maximum statistics ....

G. E. P. Box, W. G. Hunter, and J. S. Hunter, Statistics for experimenters: An Introduction to Design, Data Analysis, and Model Building,John Wiley & Sons, 1978.

....an experiment should be carried out. The simplest case is a set of fully described runs, that is, an instance of each application run with each combination of the controlling variables specified values declared in the experiment. Designing such an experimental plan follows a well known process [BOX78], widely used in most scientific disciplines. Based on the requirements (e.g. standard error, interactions) such techniques define the optimal set of runs that match the requirements at the lowest cost (in terms of time and resources) A number of plan design plug ins can drive this package. Our ....

George E.P. Box, William G. Hunter, "Statistics for Experimenters, an introduction to design, data analysis, and modeling building", 1978.

....is, the estimated model will reflect closed loop behaviour. The closed loop system may behave very differently to the open loop system. These observations become more evident when one considers figs. 2 and 3. 4 Implications of use of closed loop data in statistical modelling are examined in [6]. G u y H Figure 2 A closed loop system in a black box situation: only u and y are observable Fig. 2 shows a system consisting of two components, the productive part, represented by G , and a feedback control H, representing, for example, management controls. As mentioned above and ....

..... number of hierarchy levels involved . size, structure, location and physical cultural distance between teams involved . control procedures, policies and other mechanisms . its degree of automation or computer support . alignment between the agents involved, for example, degree of model clash [6] and . structure and other attributes of the global process and the information network involving developers, users, supporters, marketeers, their managers and others. 6 5 The global software process is seen as the process that encompasses the activity not only of developers and maintainers ....

Box GEP, Hunter WG and Hunter JS, Statistics for Experimenters - An Introduction to Design - Data Analysis and Model Building, Wiley, New York, 1978

....data transformation. Grize (1991) extended the procedure to unreplicated designs by adapting the variance estimation for the coefficients. A description of the Plot procedure and different variance estimation methods follows in Section 2. On the other hand the mean variance plot (compare e.g. Box, Hunter and Hunter, 1978) has been used for Taguchi experiments by Logothetis (1990) resulting in the fi Method. A generalization suggested by Engel (1992) which as well as the original fi Method possibly leads to inconsistent estimates for the transformation parameter , has been adapted to model (2) by Kunert and ....

.... g( y ) ffi fi y ; 15) then the transformation attained by this procedure will be proportional to the Box Cox transformation stated in equation (1) T c ( y ) 8 : c ffi(1 Gamma fi) 1 Gammafi y : fi 6= 1 c ffi ln( y ) fi = 1 9 = T ( y ) with = 1 Gamma fi (compare also Box, Hunter and Hunter, 1978). This motivates another common way to identify an appropriate data transformation based on the estimation of the parameter fi using equation (15) By taking the logarithm we get ln oe y = ln ffi fi ln y and fit the linear model ln (S y ) ln ffi fi ln (y) e 3 ; e 3 (0; oe 2 3 I) 16) ....

Box, G. E. P., Hunter, W. G., and Hunter, J. S. (1978). Statistics for Experimenters -- An Introduction to Design, Data Analysis and Model Building. Wiley series in probability and mathematical statistics.

....(Figure 6) The training points are easily identified as corresponding to the extremes of the distribution relative to the new rotated axis. Design of experiment theory helps to determine the optimal set of training points for the simulations in such a way that the best model is generated [10,11]. The most straightforward way is to perform the simulations with all possible combinations of the extremes of block performances. This approach is called full factorial analysis and requires a full set of 2 n simulations. It is possible to reduce this number. Fractional factorial designs ....

G. E. P. Box, W. G. Hunter, J. S. Hunter, Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building, John Wiley & Sons, 1978.

....models is that they usually have parameters that are difficult to measure, and, in the end, assumptions must be made in order to set those parameters. It is often easier to derive a new model that empirically describes the behavior of the benchmark. There are several papers describing this process [BHH78, Sch93]. Whatever method is used, one must be sure to validate the model over the range of inputs that are expected by the overall program. This is important for the accuracy of the global model. If a benchmark model is inaccurate at any point in this range, it will effect the accuracy of the global ....

Box, G.E.P., Hunter, W.G. and S. Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. Wiley Series in Probablility and Mathematical Statistics. John Wiley and Sons, Inc., NY, 1978.

....of the data point. Important design issues for the squashing algorithm include the choice of the number and location of the points in the parameter space to evaluate the likelihoods at, and the number of squashed data points to ensure a sufficiently good approximation. Various factorial designs [3] in the parameter space are suggested in [7] While this method is universal, as we noted above, for SVMs we can sample the values of w from the prior distribution in equation 4. The choice of the intercept term b can be made from the condition that the hyperplane goes through the cloud of ....

G. Box, W. Hunter, and J. Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. John Wiley & Sons, New York, 1978.

....doesn t use memory. 5. MRS: same as MRB, but uses only semantic features. 6. MRP: same as MRB, but uses only pictorial features. Six rst time PicHunter users, naive as to the experimental purposes, participated in this study. They ran the experiment in a 6 users 6 versions Latin square design [33]. Each user went through 15 target searches, terminating the search under the target testing paradigm; all searches terminated successfully. The results of these experiments are shown in Table I. The rst row has the average number V of 9 image displays visited before convergence to the target; ....

....V D. All 7 versions were run with the same set of 15 target images, which was di erent from the set of 15 images of the experiments of section V D. 7 users, who were naive as to the purposes of the experiment and had never used PicHunter before, participated in the 7 7 Latin square design [33]. The results are shown in Table V, which uses the same notation as that of Tables I and II. The same two experienced users who participated in all the previous experiments also ran a subset of the experiments. The user model in the new version of PicHunter (the results of which are shown in the ....

G. E. P. Box, W. G. Hunter, and J. S. Hunter, Statistics for Experimenters: An Introduction to Design, Analysis, and Model Building, J. Wiley and Sons, New York, N.Y., 1978.

....development cycle is kept to a minimum, a critical aim of all semiconductor maufacturers. Statistical methods, known as design centering or response surface methodology, are used to reduce the number of simulations required, in order to arrive at optimum settings for the manufacturing equipment [1,2]. Thus a set of input factors are varied and their effect on some output responses are analysed. Using RS 1 1 a set of experiments is designed, containing different settings for each control factor. There may be upwards of 50 separate simulations that need to be carried out to determine the ....

....EXPERIMENTAL DESIGN Response Surface Methodology (RSM) techniques have been employed in order to reduce the number of simulations that are required to give the necessary information about a particular process. As Box states, these techniques are essentially addressing the following questions [1]: ffl How is a particular response affected by a given set of input variables over some specified region of interest ffl What settings, if any, of the inputs will give a product simultaneously satisfying desired specifications ffl What values of the inputs will yield a maximum for a specific ....

G.E.P.Box and W.G.Hunter and J.S.Hunter. Statistics for experimenters : An introduction to design, data analysis and model building New York: Wiley 1978.

....issues for the squashing algorithm include the choice of the number and location of the points in the parameter space to evaluate the likelihoods at, and the number of squashed data points to ensure a sufficiently good approximation. Paper [MRD ] suggests to 5 use various factorial designs [BHH78] in the parameter space. While this method is universal, as we noted above, for SVMs we can sample the values of w from the prior distribution in equation 6. The choice of the intercept term b can be made from the condition that the hyperplane goes through the cloud of training points. We ....

G.E.P. Box, W.G. Hunter, and J.S. Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. John Wiley & Sons, New York, 1978.

....on our performance measures. Rather, we want to see the general structure in the experimental data, the location of any maxima or trends in the performance measures, to point in an appropriate direction for real user testing. Response surface techniques, modified for discrete data, are appropriate (Box et al. 1978) . Figure 8 shows how mean duration varies with L c and L o . The mean duration of tasks is lowest with a larger capture range for the OK button and a small or zero capture range for the Cancel button. This is partly an effect of the relative distribution of clicking tasks; confirmation is 20 ....

Box, George E. P.; Hunter, William G.; and Hunter, J. Stuart 1978. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. John Wiley & Sons, Inc.

....with missing values described elsewhere [25,26] There are a number of limitations in the modelling effort which arise from the use of historical, unplanned data, that is . data arising from continuous operations and not from a designed experiment. 13] Some of the dangers are listed below [8,13]: Error in the model may not be random but a result of the joint effect of several variables not included in the model (possibly, not even measured) An observed false effect of a visible variable may be caused by an unmeasured latent variable. If this changes the model may yield misleading ....

Box GEP, Hunter WG and Hunter JS, Statistics for Experimenters - An Introduction to Design - Data Analysis and Model Building, Wiley, New York, 1978

....the 95 confidence interval is defined as the interval within which the true mean value of the population will lie with probability 0. 95, and is computed from the t distribution [4] Using a t test, two experiments can be compared based on their sample means [4] or on their confidence intervals [5]. Exp. No. No. of Subjects x Sx Mode 95 confidence interval 1 30 3.13 2.06 3 [2.376,3.890] 2 10 4.00 2.31 5 [2.372,5.627] 3 10 3.20 2.15 2,3 (tie) 1.169,4.715] All 50 3.32 2.11 2,3 (tie) 2.720,3.920] Female 30 3.07 2.43 1,6 (tie) NA Table 2. The statistics for three different ....

....10 subject test, and a t test of the means reveals that there is not a significant difference of the means at a 95 confidence level. All experiments are consistent based on their confidence intervals, and this measure of correspondence provides a stronger measure than simply comparing the means [5]. The numbers chosen by the subjects can be translated into amplitudes of the noise function and related to the peak to peak amplitudes of joint rotations. For the male running simulation, the largest joint rotation is the y rotation of the shoulder (from Table 1) The y shoulder trajectory with ....

[Article contains additional citation context not shown here]

BOX, G. E. P., HUNTER, W. G., AND HUNTER, J. S. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. John Wiley, New York, 1978.

....is to obtain inexpensive and accurate estimates of the performance values in lieu of expensive circuit simulations. The savings provided by these models can, however, be offset if the construction of these performance models from simulation data is itself very expensive. Experimental designs [3] and response surface modeling [4] provide a systematic and computationally attractive means of obtaining the circuit performance models. For statistical circuit design, various performance modeling approaches have been proposed in the past. Yu et al. 5] adopted the assumption that all MOSFET ....

G. E. P. Box, W. J. Hunter, and J. S. Hunter, Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. New York:Wiley, 1978.

....to compute the effect of each factor, we can directly use the definition of effect. For example, the effect of factor x 1 on response y in Table 3. 2 is eff 1 = 2 8 (y 5 y 6 y 7 y 8 Gamma y 1 Gamma y 2 Gamma y 3 Gamma y 4 ) Estimated effects can also be obtained using Yate s algorithm [3, 4]. Usually we are only interested in the main effect and two factor interaction effect. In most of the applications, higher order interactions are very small. Main effect The difference in average process performance between the high and low settings of an experimental factor. Two factor ....

....efficient experiments, we want to study as many experimental factors as we can with the fewest observations. A common strategy in 2 A two level design has resolution R if R is the smallest number of factors in any contrast in the group of defining contrast. For details about resolution, see [3]. analyzing 2 k Gammap design is to identify those active factors whose main effects can be associated with more significant importance, discounting the possibility of interactions at the beginning. We assume that, the two factor interactions are not likely to be large in comparison to the ....

[Article contains additional citation context not shown here]

G. E. P. Box, W. G. Hunter, and J. S. Hunter, Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building, John Wiley and Sons, Inc., 1978.

....the accuracy of the resultant macromodel is sacrificed. Hence, there is a direct tradeoff between the cost of generating the model versus the accuracy of the model. Experiments can be generated statically or dynamically. Static experiments are either designed using factorial design techniques [11] or done manually using previous knowledge of the target space. On the other hand, dynamic techniques use no previous knowledge of the target space, but adapt to provide the best tradeoff between cost and accuracy. In the following paragraphs, we present the four primary aspects of our ....

....they are selected to be within the feasible input domain. Assuming that each variable has been normalized such that its low and high values are Gamma1 and 1 respectively, then a full twolevel factorial plan with n variables requires 2 n experimental runs for all possible combinations [11]. The number of experimental runs can be reduced substantially by using a 2 n Gammap III fractional factorial plan. Interested readers are referred to [4, 11] for more details. 3.2. Variable Screening and Grouping As discussed earlier, not all the input variables have the same effect on the ....

[Article contains additional citation context not shown here]

G. Box, W. Hunter, and J. Hunter, Statistics for Experimenters: an Introduction to Design Data Analysis and Model Building. John Wiley, 1978.

....the source data. ffl It may be beneficial to apply measures of information gain [Qui86] such as entropy, when determining which two attributes to select for segmentation or for the optimal threshold values for support and confidence. ffl The technique of factorial design by Fisher [Fis60, BHH78] can greatly reduce the number of experiments necessary when searching for optimal solutions. This technique can be applied in the heuristic optimizer to reduce the number of runs required to find good values for minimum support and minimum confidence. Other search techniques such as simulated ....

G. E.P. Box, W.G. Hunter, and J. Stuart Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. John Wiley and Sons, 1978.

....process simulator, PDFAB, is a combination of device and process simulators capable of performing both deterministic and statistical simulations [25] Simulations are comprised of generating 2. Fractional Factorial and Latin Hypercube as well as other deterministic experiments are described in [3]. 26 process parameters using random number generators, evaluating components for subsequent process steps, and then calculating the IC parameters and parametric data measurements using element models analytically [22] Deterministic parameters are generated by setting the mean to the value and ....

.... case 4: if (strcmp(argv[1] d ) printf( Usage: s [ e expname] a] c creator] d from date to date] n ,argv[0] exit(0) strcpy(pred, timecreated ) dt = sys.string to time(argv[2] sprintf(date, d ,dt) strcat(pred,date) strcat(pred, AND timecreated ) dt = sys.string to time(argv[3]) sprintf(date, d ,dt) strcat(pred,date) break; default: printf( Usage: s [ e exname] a] c creator] d from date to date] n ,argv[0] exit(0) Open the file if ( fp=fopen( string.file , w ) NULL) printf( CANNOT OPEN FILE n ) exit(0) Open the Database testDB = new ....

G. E. P. Box, W. G. Hunter, and J. S. Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. John Wiley and Sons, Inc., 1978.

....repeats itself until the error is sufficiently small. Variable screening and grouping: Through systematic experimental runs the significance of each variable, i.e. its effect on the output response, is estimated. Experimental runs are performed using the 2 n Gammap III fractional factorial plan [5]. The variables below a certain threshold level are neglected. Selected variables are further grouped into layers with the more significant variables in the upper layers. This classification reflects the varied influences of different variables on the output response and has a significant impact ....

G. Box, W. Hunter, and J. Hunter, Statistics for Experimenters: an Introduction to Design Data Analysis and Model Building. John Wiley, 1978.

....knowledge [19] and its increasingly statistical bent, the literature of Machine Learning (ML) contains only a few papers [3] 14] that consider software for induction as a tool to assist these shared goals. In real applications a long iterative process of experimentation and data analysis [2] is almost always necessary [8] 6] This paper describes tools and techniques for the induction and presentation of sets of decision rules [7] to help that process. The next section describes a thought experiment that will help explain and clarify the phenomena investigated here. Section 30.3 ....

....28 2.7 14.3 [17,19] 9, h d 22 1.8 16.7 [14,18] 9,9] d 20 1.0 20.0 ( 9] n [9,9] d 3 0.4 25.0 [4,8] y [10, h d 9 0.8 12.5 [15,19] 1] s d 6 0.7 28.6 ( 8] n ( 1] d consequent totals 7.4 17.4 p 2 13.7 32.9 y p 10 4.9 26.0 [20, s p 29 7. 8 62.5 [17, 2, s p 12 37.1 61.8 ( 19] [2,8] p 27 22.4 73.4 ( 16] p consequent totals 85.9 58.2 d default 6.7 13.0 Classifier Totals 100.0 52.2 TABLE 30.2. A ruleset built by the unmodified C4.5 (upper part) and a ruleset built to have a high proportion classified positive (lower part) Jason Catlett example: a more complex classifier ....

George E. Box, William G. Hunter, and J. Stuart Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis and Model Building. Wiley, 1978.

No context found.

G. E. P. Box, W. G. Hunter, and J. S. Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. Series in Probability and Mathematical Statistics. Wiley, New York, NY, USA, edition, 1978.

No context found.

G. E. P. Box, W. G. Hunter, and J. S. Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. Wiley, New York, first edition, 1978.

No context found.

G. E. P. Box, W. G. Hunter, and J. S. Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. John Wiley & Sons, 1978.

No context found.

G. Box, W. Hunter, and S. Hunter, Statistics for Experimenters: an Introduction to Design, Data Analysis, and Model Building, 1st ed., John Wiley & Sons, New York, NY, 1978.

No context found.

G.E.P Box, W.G. Hunter, and J.S. Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. John Wiley and Sons, New York, 1978.

No context found.

Box GEP, Hunter JS, Hunter WG. Statistics for Experimenters: An Introduction to Design, Data Analysis and Model Building. Wiley: New York, 1978.

No context found.

Box, G.E., Hunter, J.S., and Hunter, W.G., Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. (John Wiley & Sons, 1978).

No context found.

G. E. P. Box, W. G. Hunter., and J. S. Hunter, Statistics for experimenters: an introduction to design, analysis, and model building. John Wiley & Sons, New York, 1978.

No context found.

G. E. P. Box, W. G. Hunter, and J. S. Hunter. Statistics for experimenters: an introduction to design, analysis and model building. John Wiley & Sons, New York, 1978.

No context found.

Box, G E P, Hunter, W G & Hunter, J S, 1978, Statistics for Experimenters: An Introduction to Design, Data Analysis and Model Building, Wiley Series in Probability and Mathematical Statistics.

No context found.

G.E.P. Box, W.G. Hunter, and J.S. Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis and Model Building. Wiley-Interscience, 1978.

No context found.

G. E. P. Box, W. G. Hunter, and J. S. Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. John Wiley & Sons, 1978.

No context found.

George E. P. Box, William G. Hunter, and J. Stuart Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. John Wiley & Sons, 1978.

No context found.

G. Box, W.G. Hunter, and J. Stuart. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. John Wiley & Sons, New York, 1978.

No context found.

Box, G. E. P. , Hunter, W. G., and Hunter, J. S. (1978), Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building, John Wiley ans Sons, New York.

No context found.

Box, G.E.P., Hunter, W.G., and Hunter, J.S. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building. New York: Wiley and Sons, Inc., 1978.

No context found.

Box, George E.P., William G. Hunter, and J. Stuart Hunter, (1978) Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building, John Wiley & Sons, New York.

No context found.

Box, G., Hunter, W., and Stuart, J. (1978). Statistics for Experimenters: An Introduction to Design, Data Analysis and Model Building. Wiley & Sons.

First 50 documents  Next 50

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC