首页>
外国专利>
Computer-aided studying of probabilistic network from data record of measured, experimentally determined and/or empirical values, comprises studying structure of non-directed graphs having nodes and non-directed edges from the data record
Computer-aided studying of probabilistic network from data record of measured, experimentally determined and/or empirical values, comprises studying structure of non-directed graphs having nodes and non-directed edges from the data record
The method for computer-aided studying of a probabilistic network from data record of measured, experimentally determined and/or empirical values, comprises studying the structure of non-directed graphs having nodes (1, 2, 3, 4, 5, 6, 7, 8, 9) and non-directed edges between the nodes from the data record, producing non-directed sub-graphs from the non-directed graphs for each variable, and studying the structure and parameter of directed sub-graphs with nodes and directed edges between the nodes and/or the structure and parameter of sub-graphs of probabilistic graphic models. The method for computer-aided studying of a probabilistic network from data record of measured, experimentally determined and/or empirical values, comprises studying the structure of non-directed graphs having nodes (1, 2, 3, 4, 5, 6, 7, 8, 9) and non-directed edges between the nodes from the data record, producing non-directed sub-graphs from the non-directed graphs for each variable, and studying the structure and parameter of directed sub-graphs with nodes and directed edges between the nodes and/or the structure and parameter of sub-graphs of probabilistic graphic models with nodes and edges between the nodes from each non-directed sub-graph independent of the other non-directed sub-graphs. The probabilistic network comprises directed graph structure with nodes and directed edges between the nodes. The nodes represent variables of the data record and the directed edges dependencies between the variables. The dependencies are described by parameter of probability distributions. The non-directed sub-graphs comprise nodes and non-directed edges between the nodes in the environment of the respective variables. The respective directed sub-graph is learned, so that the directed sub-graph contains only nodes, which are present in the corresponding non-directed sub-graphs as nodes, and the directed sub-graph contains only directed edges, which are present in the corresponding non-directed sub-graphs as non-directed edges. For studying the structure of the non-directed graphs, a test-based learning process such as a statistical independence test and/or personal computer algorithm and/or three-phase dependency analysis algorithm is used. The test-based learning process is developed, so that variables conditional dependence of the respective variables are added to a candidate record of variables, which fulfill a given heuristic function, and variables, which are subset of variables of the candidate records giving conditional independence of the respective variables, are removed from the candidate record. The heuristic function is fixed, so that the variable is added to the next candidate record, which maximizes the smallest conditional dependence of the respective variable tested for all possible subsets at variables of the candidate record. The directed edges are produced between the respective variables and the variables of the candidate record after adding and removing the variables for the respective variable. A score-based learning process is used for learning the structure and parameter of the respective directed sub-graphs. An evaluation after the respective directed sub-graphs is searched in the score-based learning process under consideration. The score-based learning process uses greedy-algorithm after the respective directed sub-graphs for searching. A local structure is fixed within the non-directed graphs for the respective variable. The local structure as nodes comprises the respective variable, the neighbors of the respective variables and if necessary neighbor of higher degrees and the non-directed edges. The local structure of the non-directed sub-graphs represents the respective variables. After learning the respective directed sub-graph, the nodes are removed from the directed sub-graphs, which not belong to the Markov-blanket. After removing the nodes not belonging to the Markov-blanket, a feature partial directed graph is produced, by which the probabilities are determined from the directed sub-graphs for each occurring edge, in which the direction edges are directed. The edges are non-directionally arranged and/or actually no edge is present. A Bayesian network is learned. The data record comprises biological, medical and/or biomedical data such as gene expression samples, occurrence of diseases, clinical data, life-habits of patients and/or pre-existing diseases of patients. The data record comprises data from an automation system, a power generation system and/or a communication network. Independent claims are included for: (1) a method for computer-aided simulation of data based on a probabilistic network; and (2) computer program product with a program code stored on a machine-readable carrier.
展开▼