Download to read offline
Dss
- 1. DSS PRESENTATION MODEL SELECTIONAND SEQUENCING IN DECISION SUPPORT SYSTEMS GROUP MEMBERS: 116101-PUJA PUNIA 116102-RAHUL YADAV 116103-RAHUL KUMAR MEENA 116104-RAHUL SINGH RAJPUT 116105-RAJAT GOEL
- 2. ABSTRACT A crucialproblem confronting the users of decision support systems(DSS) is the identification of an appropriate model or a sequence of models that may be used to solve a particular problem. If a single model in the model base cannot satisfy user requirements, the solution procedure seeks to obtain a string of models such that some performance measure is minimized.
- 3. INTRODUCTION A complementarytrend is the popularity of decision support systems (DSS), where the intended functionality of these systems is to assist decision makers' solve semi-structured and un- structured problems that require extensive usage of computational models. The high costs of developing and maintaining models further reinforce the importance of model management in organizations. A need to maintain consistency, integrity, non-redundancy, and flexibility and the need to enhance the availability of models on an organization-wide basis, has led to the evolution of software known as model management systems (MMS).
- 4. INTRODUCTION CONTD. TheMMS component of a DSS provides decision support by abstracting procedural and technical aspects of model implementation and making these invisible to the non-technical user . Two fundamental concerns faced by a user in interacting with a DSS are: what is the appropriate model to use for a particular problem, and if such a model does not exist as a single entity, how can the models defined in the model base be combined to solve the problem? These concerns are particularly relevant when complex decision processes require the utilization of several models in sequence. The model linking procedure must not violate the fundamental objective underlying MMS that of relieving the user from procedural and technical details.
- 5. CONTD. This paperdevelops a procedure for model sequencing that permits the construction of ad hoc model chains. Knowledge is extracted from a specialist and stored in the model base, allow- ing the user's interaction with the DSS to be limited to conceptualizing the problem in terms of output requirements.
- 6. MODEL SELECTION AND INTEGRATION Model selection deals with locating an appropriate model in the model base that maps into the problem posed by the decision maker. If this search process identifies a subset of the model space that is potentially applicable, i.e. more than one model is able to satisfy the user's output requirement, then the MMS chooses to select and execute a single model from this subset. The model integration problem concerns itself with finding a target set of models to be combined and executed, if, in fact, no single model is able to provide the requisite output.
- 7. CONTD. Blanning proposeda model representation scheme where model inputs and outputs are treated as tuples in a relational table, with each tuple containing information about a specific variable, the models to which it is input, and the models that estimate it. Geoffrion describes a structured modelling approach to model representation where models are represented in canonical form. Structured modelling is an extension of graph-based approaches and the framework uses a "hierarchically organized, partitioned and attributed acyclic graph to represent a model or a model class“. The construction of composite models allows input ports to represent the input requirements of the model and output ports to represent the outputs produced. These input/output relationships are used to construct higher-level composite models.
- 8. CONTD. The useof information in our approach is parsimonious, as are our assumptions regarding the knowledge possessed by the user of the MMS (the only information required by the approach are model inputs and outputs). The model sequencing procedure provides the user with an optimal sequence of models, where the criterion for optimality can be varied if so desired.
- 9. PROCEDURE The mannerin which the problem is modelled is influenced by the availability of input data. The objective of MMS is to assist the user select an appropriate model (or a model string) from the model base. This selection may be based on either (i) desired output variables (ii) both desired output variables and input variables that can be supplied. The model sequencing procedure described here operates on an existing model base. Two assumptions are made regarding the model base: (i) it contains a large number of models represented using perhaps different formalisms, and (ii) these models have been created by knowledgeable users or specialists and have passed some type of quality assurance test to ensure their validity.
- 10. 1. MODEL BASECATALOGS Three primary catalogs are maintained in the system, one each for models (MODCAT), out- put variables (OUTCAT) and input variables (INCAT). MODCAT contains the name of the model, a brief description of its scope, and assumptions underlying its use, such as linearity and independence. The catalog of outputs is annotated with a description of the procedure by which the output is computed.
- 11. COMPONENTS OF THE MODELSELECTOR The model selector software consists of two primary modules called DIALOG and SEARCH. The former provides a user- friendly interface for a non-technical user and uses the catalogs to ascertain if a single stored model can satisfy a specific user need. SEARCH consists of a 0/1 integer programming formulation that seeks to obtain an optimal path to the required outputs. The DIALOG component. DIALOG is menu- driven and prompts the user to supply required output variables and associated objectives.
- 12. 3. ITERATION BETWEEN SEARCHAND DIALOG DIALOG analyses the final solution obtained from SEARCH and uses the catalogs to display the list of model names, input variable names, and the associated cost of model execution. The primary objective now is to seek another model string that does not require this input variable. In this fashion, DIALOG and SEARCH are executed iteratively to obtain an acceptable model string. Iteration between SEARCH and DIALOG may also occur if either the user finds the objective(s) of the model(s) selected by the system to be inappropriate, or, the cost of executing the model string is unacceptable.
- 13. CONTD. For example,if the user does not specify an objective a priori, and the system selects an estimation model to compute the value of a certain variable, the user has the flexibility of requesting SEARCH to find another model with an optimization objective. The approach described combines management science techniques and basic user interface design principles to remove procedural and technical obstacles in the utilization of models by end-users.
- 14. A PROTOTYPE SYSTEM Prototype system has been applied to a model base. The menu driven DIALOG component is programmed in GWBASIC and the integer program is solved using LINDO.
- 15. ASCERTAINING COSTS Thecost parameters, and the policies used for obtaining them, can be classified into two distinct Classes. Method employed for estimating the dollar value. 1. In some instances the estimates used are specific to the computer system (and the associated charge-back rules)where the model base is installed. 2. For the prototype developed, all models are stored and executed on the IBM 3090/600E at Louisiana State University (LSU), Baton Rouge. Thus, various cost parameters were estimated based on the charging algorithm employed by LSU's System Network Computer Center.
- 16. CONTD. (I)Model execution CPUusage costs are computed as the product of actual CPU time (in seconds) and a factor F (S/second). The value of F varies depending on the time of the day. For prototype system, it was assumed that all models will be executed during "demand time" (i.e. between 8 am and 5 pm). After a model had been debugged and validated, average CPU usage time was calculated by executing each model with historic (or representative) data.
- 17. (a)Primary inputs availablein the database cost was obtained by employing the university's charging algorithm for performing I/O operations. Cost parameters were calculated by multiplying the average number of I/Os required by the model and a charge factor I ($ per 100 I/Os). Standard tape mounting charges used by the university were incorporated for obtaining the cost of inputs that required reading data from magnetic tapes.
- 18. (b)Primary inputs suppliedby decision maker Some primary inputs (e.g. expected growth rate) must be supplied by the decision maker, based on his/her experience or intuition, the cost of these inputs were arbitrarily set to a low value.
- 19. C)Primary inputs availablefrom 'outside‘ sources Primary inputs available from 'outside‘ sources fall into two major categories. 1.Inputs are obtainable from commercial on-line databases. The cost associated with this type of input is based on the connect time to the on-line database. These inputs can be further classified into two classes: (a) input data that can be stored for future use, and (b) up-to-date, current data that must be down-loaded from the database each time the input is required.
- 20. CONCLUSION Model manipulationcontinues to be an important issue for both academics and practitioners. Model management systems attempt to enhance utilization by effectively shielding non-expert users from internal representations of models in computers. Our research has addressed one respect of model management that concerns itself with model selection and sequencing. In this paper we have described mechanisms for achieving model selection and sequencing that are independent of model implementation and storage.