JP5017939B2 - Causal structure acquisition apparatus, causal structure acquisition method, causal structure acquisition program, and computer-readable recording medium recording the same - Google Patents

Description

  The present invention relates to a causal structure acquisition apparatus, a causal structure acquisition method, a causal structure acquisition program, and a computer-readable recording medium in which the causal structure acquisition program supports the creation of a causal structure of knowledge information used in an inference system such as diagnosis and prediction. It is.

  In an inference system using a causal structure, “graphical modeling”, “SGS algorithm”, “WL algorithm” (Non-patent Documents 1 and 2) and the like are often used as methods for acquiring the causal structure itself.

Further, Patent Document 1 describes the knowledge information of the structural unit of the target device associated based on the mechanical structure and the knowledge information of the process of the structural unit related based on the linkage of the operation of the structural unit, and the configuration when the process is executed. There has been disclosed a knowledge information processing apparatus capable of inputting knowledge information with a simple expression by storing knowledge information of materials propagating between units in a knowledge base.
Japanese Patent Laid-Open No. 5-027980 (Publication date: February 5, 1993) Masami Miyagawa “Statistical Library Graphical Modeling” Asakura Shoten 1997 Masakawa Miyagawa “Statistical Causal Reasoning” Asakura Shoten 2004

  However, if the above-described algorithms such as “graphical modeling”, “SGS algorithm”, and “WL algorithm” are to be used as they are, a very large amount of data is necessary (about several thousand), which is not practical. Moreover, depending on the setting of the condition for deleting the arrow line indicating the causal relationship between variables and the setting of the end condition, the arrow line may remain between many variables, resulting in an unintelligible causal structure.

  Further, in the knowledge information processing apparatus of Patent Document 1, it is necessary to store in advance the knowledge of the constituent unit of the target device, the knowledge of the process, and the knowledge of the material in the knowledge storage unit. Preparing is not practical.

  The present invention has been made in view of the above problems, and its purpose is to provide a causal structure acquisition apparatus and causal structure that can easily acquire the causal structure of knowledge information used in an inference system such as diagnosis and prediction. An acquisition method, a causal structure acquisition program, and a computer-readable recording medium on which the acquisition method is recorded are realized.

  In order to solve the above problems, the causal structure acquisition apparatus according to the present invention acquires a plurality of sets of measurement data of two variables selected from a variable group including a plurality of variables from the measurement data storage unit, and acquires the acquired data. Based on the measured data, the test statistic for the independence of the two variables is obtained. If the test statistic is greater than the upper threshold, there is a causal relationship between the two variables. The test statistic is the lower threshold. If it is smaller, there is no causal relationship between the two variables, and if the test statistic is between the two thresholds, between the variables that determine whether there is a causal relationship between the two variables It is characterized by comprising a relationship generation means.

  Also, the causal structure acquisition method according to the present invention is a causal structure acquisition method by a causal structure acquisition apparatus, wherein the inter-variable relationship generation means provided in the causal structure acquisition apparatus is selected from a variable group including a plurality of variables. Obtain multiple sets of measurement data of two variables from the measurement data storage unit, obtain the test statistic of independence of the two variables based on the acquired measurement data, and if the test statistic is greater than the upper limit threshold For example, there is a causal relationship between the two variables. If the test statistic is smaller than the lower threshold, there is no causal relationship between the two variables. If the test statistic is between the two thresholds, the It is characterized by including an inter-variable relationship generation step for determining that there is no causal relationship between the two variables.

  According to the above configuration, the test statistic for the independence of the two variables is obtained based on the measurement data of a plurality of sets of two variables selected from a variable group including a plurality of variables, and the test statistic is determined in advance. By comparing with the two thresholds (upper limit threshold, lower limit threshold), the state of the causal relationship between the two variables (whether there is a causal relationship / no causal relationship / whether there is a causal relationship) is determined. To do.

  Thereby, since the thing with a big test statistic can be automatically determined as two variables which have a causal relationship, the presence or absence of the causal relationship between the variables contained in a variable group can be determined easily. In addition, since two threshold values of the upper limit and the lower limit are used, it is possible to determine a state in which it is not known whether there is a causal relationship. Therefore, it is possible to determine whether a state where it is not known whether there is a causal relationship is a causal relationship / no causal relationship based on examination from a viewpoint other than the test statistic. Therefore, it is possible to easily acquire a causal structure of knowledge information used in an inference system such as diagnosis and prediction.

  Furthermore, in the causal structure acquisition apparatus according to the present invention, the inter-variable relationship generation unit sets a variation threshold smaller than the lower limit threshold, and the test statistic is between the variation threshold and the lower limit threshold. While determining whether or not there is a causal relationship between two variables, there is a causal relationship between the two variables of interest either directly or indirectly through another variable. Alternatively, the variation threshold value is reduced until it is determined that the causal relationship is unknown.

  According to the above configuration, the two variables of interest have a causal relationship or are not connected by connection between variables in a state where it is not known whether there is a causal relationship, that is, for example, from a control variable to an objective variable. If there is no route to, the most probable route can be created by reducing the threshold value for determining causality. Therefore, examination of the causal structure becomes easy.

  Furthermore, in the causal structure acquisition apparatus according to the present invention, the inter-variable relationship generation unit changes the presence / absence of the determined causal relationship according to causal relationship defining information that predefines the presence / absence of a causal relationship of another variable for each variable. It is characterized by doing.

  According to the above configuration, if the connection relationship with other variables is known in advance as the causal relationship defining information, the presence or absence of the causal relationship can be changed by using the information.

  Furthermore, in the causal structure acquisition apparatus according to the present invention, the inter-variable relationship generation means arranges variables located upstream and variables located downstream in rows and columns in a route connecting variables having causal relationships. A variable relation table in which the presence or absence of a causal relationship between two variables specifying the element is set for each element of the two-dimensional matrix, and the row and column variables are set in the same order with respect to the rows and columns, and It is characterized by rearranging in the order of precedence.

  According to the above configuration, in the inter-variable relationship table sorted as described above, if the presence / absence of the causal relationship is accurately set, the elements having no causal relationship gather in the upper right (or lower left) (FIG. 13). ). Therefore, it is possible to easily check the validity of the setting of the inter-variable relationship table by examining whether or not there is an element that is set as having a causal relationship in an area where elements having no causal relationship are gathered.

  Furthermore, the causal structure acquisition apparatus according to the present invention is characterized by comprising display means for displaying variables included in the variable group together with a connection line that connects two variables having a causal relationship.

  According to said structure, the variable of a variable group can be further connected and displayed by a connection line according to a causal relationship. Therefore, it becomes possible to easily understand the causal structure of the variable group.

  Furthermore, in the causal structure acquisition apparatus according to the present invention, the display means displays only the variable of interest and the variable that has a causal relationship directly with the variable of interest or indirectly through another variable. It is characterized by being displayed together with the connection line.

  According to the above configuration, in addition to displaying only the variables of interest and variables that are causally related directly to the variables of interest or indirectly through other variables, together with the connection lines, the variables It is possible to easily understand the causal structure related to the variable of interest in the group.

  The causal structure acquisition apparatus may be realized by a computer. In this case, the control program for the causal structure acquisition apparatus that causes the computer to operate the causal structure acquisition apparatus by operating the computer as each of the above means. (Causal structure acquisition program) and a computer-readable recording medium on which it is recorded also fall within the scope of the present invention.

  As described above, the causal structure acquisition device according to the present invention acquires a plurality of sets of measurement data of two variables selected from a variable group including a plurality of variables from the measurement data storage unit, and acquires the acquired measurement data. Based on this, the test statistic for the independence of the two variables is obtained. If the test statistic is greater than the upper threshold, there is a causal relationship between the two variables. If the test statistic is smaller than the lower threshold, There is no causal relationship between the two variables, and if the test statistic is between two thresholds, a variable relationship generating means for determining that there is no causal relationship between the two variables. It is the structure provided with.

  Also, the causal structure acquisition method according to the present invention is a causal structure acquisition method by a causal structure acquisition apparatus, wherein the inter-variable relationship generation means provided in the causal structure acquisition apparatus is selected from a variable group including a plurality of variables. Obtain multiple sets of measurement data of two variables from the measurement data storage unit, obtain the test statistic of independence of the two variables based on the acquired measurement data, and if the test statistic is greater than the upper limit threshold For example, there is a causal relationship between the two variables. If the test statistic is smaller than the lower threshold, there is no causal relationship between the two variables. If the test statistic is between the two thresholds, the This is a method including an inter-variable relationship generation step for determining that there is no causal relationship between two variables.

  Therefore, since a variable having a large test statistic can be automatically determined as two variables having a causal relationship, the presence or absence of a causal relationship between variables included in the variable group can be easily determined. In addition, since two threshold values of the upper limit and the lower limit are used, it is possible to determine a state in which it is not known whether there is a causal relationship. Therefore, it is possible to determine whether a state where it is not known whether there is a causal relationship is a causal relationship / no causal relationship based on examination from a viewpoint other than the test statistic. Therefore, there is an effect that it is possible to easily acquire the causal structure of the knowledge information used in the inference system such as diagnosis and prediction.

  One embodiment of the present invention will be described below with reference to FIGS.

  FIG. 1 is a functional block diagram showing an outline of the configuration of a causal structure acquisition apparatus 10 according to the present embodiment. FIG. 2 is an explanatory diagram showing an outline of functions of the causal structure acquisition apparatus 10.

  The causal structure acquisition apparatus 10 is an apparatus that supports acquisition of an effective causal structure that is easily applied to data and easy to understand from the measurement data and prior knowledge.

  Specifically, as shown in FIG. 2, the causal structure acquisition apparatus 10 includes, for example, variables (printing pressure, printing speed, solder viscosity, temperature, solder amount, heating time, and the like) that are a plurality of measurement items of a certain quality inspection. It has the function of determining the presence or absence of a causal relationship between variables from the measurement data of the fillet area) and displaying the causal variables connected by arrows.

The causal structure acquisition apparatus 10 particularly has the following functions.
(1) Obtain the test statistic of independence of two variables. If the test statistic is larger than the upper threshold, there is a causal relationship between the two variables. If the test statistic is smaller than the lower threshold, There is no causal relationship between two variables, and if the test statistic is between two thresholds, it is determined that there is no causal relationship between the two variables, and the degree of causal relationship Displays the arrow of the line type corresponding to.
(2) Using the nature of the type of variable, delete extra arrows.
(3) When there is no route to the target variable of interest, the threshold value (variation threshold value) is decreased until a route is created.
(4) The causal structure is displayed by changing the line type (solid line, dotted line, line thickness, etc.) according to the magnitude of the test statistic.

  In addition, in the causal structure acquisition apparatus 10 of this Embodiment, if there exists about several 10 to several hundreds measurement data with respect to one variable group, a causal structure can be determined with sufficient precision.

  Next, the function of the causal structure acquisition apparatus 10 will be described in detail. As shown in FIG. 1, the causal structure acquisition apparatus 10 includes an inter-variable relationship generation unit (inter-variable relationship generation unit) 11, an arrow line determination unit 12, a route presence / absence determination unit 13, and a causal structure output editing unit (display unit) 14. , A data input unit 15, a type-specific variable storage unit 21, a preceding relationship storage unit 22, and a measurement data storage unit 23.

  The inter-variable relationship generation unit 11 acquires a plurality of sets of measurement data of two variables selected from a variable group including a plurality of variables from the measurement data storage unit 23, and based on the acquired measurement data, the two variables If the test statistic is larger than the upper threshold, there is a causal relationship between the two variables. If the test statistic is smaller than the lower threshold, the two variables are If there is no causal relationship, and the test statistic is between two thresholds, it is determined that it is not known whether there is a causal relationship between the two variables.

  Further, the inter-variable relationship generation unit 11 sets a variation threshold smaller than the lower limit threshold, and a case where the test statistic is between the variation threshold and the lower limit threshold is a causal relationship between the two variables. It is determined whether it is in a state where it is unknown whether or not there is a causal relationship between the two variables of interest, directly or indirectly through other variables, or whether or not there is a causal relationship The variation threshold is decreased until it is determined that the state is present.

  In addition, the inter-variable relationship generation unit 11 determines the above according to the causal relationship defining information (variable classification associated with the type and nature of the variable) that predefines the presence or absence of the causal relationship with other variables for each variable. Change the presence or absence of the causal relationship.

  The arrow line determination unit 12 determines the line type of the arrow line based on the test statistic of the independence of the two variables. Note that the shape and line type of the connecting line that connects the two variables are not limited to arrow lines, solid lines / dotted lines, and the like as long as they express the degree of causal relationship between the variables, and can be changed as appropriate.

  The route presence / absence determination unit 13 determines whether or not the variables of interest are directly or indirectly connected via other variables.

  The causal structure output editing unit 14 displays the variables included in the variable group together with an arrow line (connection line) connecting two variables having a causal relationship. In addition, the causal structure output editing unit 14 displays only the variables of interest and the variables that are causally related to the variables of interest directly or indirectly through other variables, along with arrow lines (connection lines). indicate.

  The data input unit 15 acquires variables, preceding relationships between variables, and measurement data from the user or the outside, and stores them in the variable storage unit 21 for each type, the preceding relationship storage unit 22, and the measurement data storage unit 23, respectively.

  As the type-specific variable storage unit 21, the preceding relationship storage unit 22, and the measurement data storage unit 23, a hard disk, a memory, or the like can be used.

  FIG. 3 is an explanatory diagram showing a specific example of a variable item name table stored in the preceding relationship storage unit 22. As shown in FIG. 3, the prior relationship storage unit 22 stores variable item names (printing pressure, printing speed, heating time, temperature, solder viscosity, solder amount, fillet area) in a table format. The data structure of data stored in the type-specific variable storage unit 21 and the measurement data storage unit 23 will be described later.

  Next, FIG. 4 is a flowchart showing an outline of processing in the causal structure acquisition apparatus 10.

  As shown in FIG. 4, the user inputs variables, variable classifications, and precedence relationships via the data input unit 15 (S11, S12, S13). The variable and variable classification data are stored in the type-specific variable storage unit 21. Further, the preceding relationship data is stored in the preceding relationship storage unit 22. The measurement data is stored in advance in the measurement data storage unit 23 via the data input unit 15. However, the data input unit 15 may acquire various data such as variables, variable classifications, precedence relationships, and measurement data from user input or may be configured to acquire from an external device. .

  Here, FIG. 5 is an explanatory diagram of the types of variables (causal relationship defining information) handled by the causal structure acquisition apparatus 10. FIG. 6 is a screen display example for the user to input variables and variable classifications. 6 is displayed under the control of the data input unit 15.

  As shown in FIG. 5, there are three types of variables handled by the causal structure acquisition apparatus 10: control variables, covariate variables, and intermediate / objective characteristic variables.

  The control variable is a parameter that can be adjusted by the user. Since the control variable is the variable that appears first in the preceding relationship (the order in which the variables appear), there is no arrow line coming from another variable.

  Covariate variables are not affected by intermediate / target characteristic variables. Therefore, covariate variables do not have an arrow line coming from an intermediate / target characteristic variable. As a result, the generation of a route to go around can be prevented.

  The intermediate / objective characteristic variable is an unconstrained variable and follows any preceding relationship.

  In addition, the user inputs a variable classification using an input screen as shown in FIG. In this input screen, there are provided a variable list and a classification input field for setting a variable displayed in the variable list as one of a control variable, a covariate variable, and an intermediate / target characteristic variable. In this classification input field, variables are arranged horizontally when there is a preceding relationship, and vertically when there is no relationship. That is, it is possible to simultaneously set the variable precedence relationship and the variable type classification.

  Subsequently, in step S14 of FIG. 4, the causal structure acquisition apparatus 10 determines the causal structure from the data, and creates a variable relation table (described later). Details of the processing in step S14 will be described later.

  Next, the causal structure output editing unit 14 receives a correction for the arrow line of the displayed causal structure from the user, and corrects the inter-variable relationship table (S15). And a process is complete | finished by the input of completion | finish by a user (S16).

  FIG. 7 is a flowchart showing details of the process for determining the causal structure from the data (S14 in FIG. 4).

  As described above, in the causal structure acquisition apparatus 10, the inter-variable relationship generation unit 11 acquires a plurality of sets of measurement data of two variables selected from a variable group including a plurality of variables from the measurement data storage unit 23, A test statistic for the independence of the two variables is obtained based on the acquired measurement data, and if the test statistic is greater than the upper threshold, there is a causal relationship between the two variables. If it is less than the lower threshold, there is no causal relationship between the two variables, and if the test statistic is between the two thresholds, it is determined that there is no causal relationship between the two variables. To do. Specifically, in the example of the flowchart of FIG. 7, a solid line is drawn between variables having a test statistic of 2σ (upper threshold) or more (high correlation and reliable connection), and the test statistic is 1σ. Draw a dotted line between the variables (lower threshold) and less than 2σ (upper threshold) (the correlation is low and it is not known whether they are connected), and between the other variables (less than 1σ (lower threshold)) Is set in the variable relation table (FIG. 9A) so as not to draw an arrow line.

  In order to obtain the test statistic, first, a correlation coefficient (partial correlation coefficient) between the two variables is obtained based on the measurement data of the two variables. Next, the test statistic is calculated according to the following formula 1 based on the correlation coefficient (partial correlation coefficient) between the two variables (S21).

  Next, the standard deviation of the test statistic is calculated (S22). Since the statistical test amount follows the t distribution, the standard deviation is calculated according to Equation 2 (S21).

  Here, FIG. 8 is an explanatory diagram for explaining a method in which the inter-variable relationship generating unit 11 determines the degree of the causal relationship between the variables from the test statistic.

  Then, the arrow line determination unit 12 determines the line type of the arrow line according to which range of the test statistic-line type relationship is set in advance. In the example of FIG. 8, a solid line is drawn between variables whose test statistic is 2σ or more (S23, S24), and a dotted line is drawn between variables whose test statistic is 1σ or more and less than 2σ (S25, S26). It is shown that settings are made so that no arrow line is drawn between the variables (S27). The arrow line determination unit 12 determines the line type of the arrow line between all variables (S28). Note that the upper and lower threshold values (1σ, 2σ) are examples, and can be appropriately changed according to measurement data and the like.

  FIG. 9 is an explanatory diagram showing a simple example corresponding to the above description, where (a) is a variable relation table, and (b) is a causal structure displayed based on the variable relation table of (a). .

  As shown in FIG. 9 (a), between test variables whose test statistics are 2σ or more, that is, “printing speed—solder amount”, “heating temperature—fillet area”, “solder viscosity—solder amount”, “solder amount—fillet area”. In the “solder viscosity-printing speed”, “◯” indicating that a solid arrow is drawn is set. In addition, between the variables whose test statistic is 1σ or more and less than 2σ, that is, “printing pressure−solder amount”, “temperature—solder viscosity”, “solder viscosity—fillet area” indicates that a dotted arrow is drawn. "Is set. In addition, “×” indicating that no arrow line is drawn is set between variables whose test statistic is less than 1σ, that is, between variables other than the above. FIG. 9B shows a causal structure displayed by the causal structure output editing unit 14 based on the setting shown in FIG.

  Here, in FIG. 9A, when it is set that there is no causal relationship between “solder amount—fillet area”, “heating time—fillet area”, and “solder viscosity—fillet area”, a solid line between these variables or The dotted arrows are not drawn. As a result, there is no route from the control variable (printing pressure, printing speed, heating time) to the target variable (fillet area).

  Therefore, the route presence / absence determination unit 13 searches for a route from the processing variable (printing pressure) of interest to the target variable (fillet area) (S29). If a line is not drawn on the route of interest (NO in S30), the arrow line determination unit 12 sets a line with a variation threshold smaller than the lower limit threshold until a line is drawn on the route of interest ( S31). Specifically, the variable satisfying this condition is set so that the arrow line determination unit 12 sets the variation threshold smaller than the lower limit threshold 1σ and draws a one-dot chain line between variables whose test statistic is greater than or equal to the variation threshold and less than 1σ. Set “▲” in between.

  In this example, the degree of the causal relationship between variables is determined using the upper limit threshold, the lower limit threshold, and the variation threshold. However, the number of thresholds can be arbitrarily selected as long as it is two or more. Further, although the degree of causal relationship between variables is distinguished by line types (solid line, dotted line, etc.), they may be distinguished by line thickness and color, or a combination thereof.

  FIG. 10A is an example of an inter-variable relationship table in which the one-dot chain line arrow is set, and FIG. 10B is a display of the causal relationship based on FIG. As shown in FIG. 10B, an arrow line is drawn to “printing pressure-fillet area” by drawing an arrow line (dashed line) having a test statistic greater than or equal to the variation threshold and smaller than 1σ. In addition, since the correlation of two variables is low and the arrow line of a dashed-dotted line cannot be used for causal reasoning, a warning (warning) is issued to the user.

  Note that the target processing variable and the target variable of interest can be arbitrarily set by the user based on the control of the causal structure output editing unit 14.

  As described above, according to the causal structure acquisition apparatus 10, it is possible to distinguish and display an arrow line connecting variables with a certain causal relationship as a solid line and an arrow line with an uncertain causal relationship as a dotted line.

  Note that the dotted arrow line may be deleted or changed to a solid line by the user in accordance with the control of the causal structure output editing unit 14. Further, a judgment index such as a correlation coefficient may be displayed together with the display of the causal structure.

  In addition, according to the number of arrow lines, increase or decrease the number of uncertain arrow lines (displayed with dotted lines), or display the specified number of arrow lines so that the causal relationship is displayed in order from the arrow line with a certain causal relationship. May be set as

  In the present embodiment, the presence or absence of an arrow line is determined based on whether the corresponding variables are independent or conditionally independent. That is, it is determined as independent if it is larger than the threshold. Conventionally, for example, a determination of 5% significance has been made, but an accurate determination cannot be made without a sufficient number of data. In other words, the distribution increases when there is little data. Therefore, in this embodiment, independent testing is not performed with one threshold value, but the upper and lower threshold values are used, and if the test statistic is larger than the upper threshold value, a causal relationship is established between the two variables. If the test statistic is smaller than the lower threshold, there is no causal relationship between the two variables. If the test statistic is between the two thresholds, there is a causal relationship between the two variables. It is determined that the state is unknown. The line type of the arrow line is determined according to the degree of causality.

  Next, processing for determining the inter-variable relationship from the type and nature of the variable will be described. This process is performed by the inter-variable relationship generation unit 11.

  FIG. 11 is an explanatory diagram showing a process of automatically deleting unnecessary arrow lines using the property depending on the type of variable. Note that the types and properties of the variables are as shown in FIG.

  In the example of FIG. 11, the arrow line that is included in the printing speed (parameter) from the solder viscosity (intermediate characteristic) is deleted by utilizing the property that the control variable does not have an arrow line. In this way, unnecessary arrow lines can be deleted by using the nature of the type of variable.

  Further, the inter-variable relationship generation unit 11 sets the entire inter-variable relationship table to some extent using the types and properties of the variables.

  FIG. 12 is an example of an inter-variable relationship table that is automatically determined from the type and nature of the variable.

  As shown in FIG. 12, the inter-variable relationship generation unit 11 sets “Δ” between variables that may be drawn with an arrow line based on the property shown in FIG. 5 when the variable type is set. “,” Is set between variables where there is no possibility that an arrow line is drawn.

  Next, FIG. 13 is an example in which the inter-variable relationship table shown in FIG. 12 is sorted in order from the variable upstream of the route, that is, in the order of control variable, covariate variable, and intermediate / target characteristic variable. .

  As shown in FIG. 13, when the variable relationship table is sorted, the variables are arranged in the order of the preceding relationship. And since there cannot be an arrow line from the child variable on the downstream side to the parent variable on the upstream side of the two connected variables, all the columns in the upper right half of the intervariable relationship table should be set to “x”. is there. As described above, by rearranging the variables in the inter-variable relationship table in the order of the prior relationship, inconsistencies in the inter-variable relationship table can be found. It should be noted that the state shown in FIG. 13 is converted by selecting the “preceding relationship order” check box on the display screen of FIG.

  Furthermore, FIG. 14 is an example in which the intervariable relationship table is modified based on prior knowledge input by the user.

  Here, as prior knowledge, information that “printing pressure and printing speed are not related to temperature” and “solder viscosity is related to solder amount” is set. At this time, the inter-variable relationship generation unit 11 corrects “printing pressure-temperature”, “printing speed-temperature” to “x”, and “solder viscosity-solder amount” to “◯” based on these prior knowledge. To do. Also, if “△” or “◯” is set in the upper right, a warning (warning) is issued.

  The prior knowledge may be set in advance by the user, or may be input while viewing the intervariable relationship table in FIG. The prior knowledge data structure can be in any format as long as the correspondence relationship with the intervariable relationship table can be acquired.

  Then, when the determined intervariable relationship table is obtained as shown in FIG. 14, the test statistic is calculated only for the variables for which “Δ” is set, and the line type of the arrow line is determined. Therefore, step S28 in FIG. 7 may be processed for all variables for which “Δ” is set in the inter-variable relationship table in FIG.

  FIG. 15 is a flowchart showing details of processing of the causal structure acquisition apparatus 10. This flowchart summarizes the processes described above.

  First, a user defines a variable, classifies the variable by type, and inputs the variable to the causal structure acquisition apparatus 10. Also, the target variable of interest is designated from the intermediate characteristics (S101).

  Next, the causal structure acquisition apparatus 10 stores the variable classified for each type of variable in the variable storage unit 21 for each type (S102).

  Next, if the user knows the preceding relationship, the user inputs it to the causal structure acquisition apparatus 10 (S103).

  Next, the causal structure acquisition apparatus 10 stores the input preceding relationship in the preceding relationship storage unit 22 (S104). And the causal structure acquisition apparatus 10 produces | generates the relationship table between variables in the variable relationship production | generation part 11 from the data of the variable memory | storage part 21 for every classification, and the data of the preceding relationship storage part 22 (S105). Then, the causal structure acquisition apparatus 10 sorts the inter-variable relationship table generated by the inter-variable relationship generating unit 11 in the order of prior relationship, and checks whether there is a contradiction (S106).

  Next, if the user knows in advance whether or not there is a relationship between variables, the variable relationship table is modified (S107).

  Next, the causal structure acquisition apparatus 10 reads the measurement data stored in the measurement data storage unit 23 by the arrow line determination unit 12 (S108). Then, the causal structure acquisition apparatus 10 extracts pairs of variables with “Δ” in the inter-variable relationship table at the arrow line determination unit 12 (S109). And the causal structure acquisition apparatus 10 calculates the test statistic between variables and its standard deviation in the arrow line determination part 12 (S110). When the test statistic is 2σ or more, the causal structure acquisition apparatus 10 is assumed to have a solid line and an arrow line. Moreover, when it is 1σ or more and less than 2σ, it is assumed that there is an arrow line with a dotted line. Further, it is assumed that there are no arrow lines below that (S111).

  Next, the causal structure acquisition apparatus 10 determines whether there is a route from the processing variable to the target variable by the route presence / absence determination unit 13 (S112). If there is no route, an arrow line having a statistical test amount of 1σ or less is added as a one-dot chain line until a route is created (S113). Then, the causal structure acquisition apparatus 10 outputs the automatically created causal structure to the causal structure output editing unit 14 (a solid arrow is a solid line, an uncertain arrow is a dotted line, etc.) (S114).

  Next, in the created causal structure, the user changes the location to be changed (if there is an arrow line displayed with a dotted line (S115)).

  Next, the causal structure acquisition apparatus 10 reflects the change result in the causal structure output editing unit 14. At this time, if there is an arrow of a one-dot chain line, it is output as a warning (warning) that the result of causal inference is not accurately output because the correlation is low (S116).

  FIG. 16A is an example of a final inter-variable relationship table obtained through the above processing, and FIG. 16B is a display of the causal structure based on FIG.

  As shown in FIG. 16A, “printing pressure−solder amount”, “printing speed—solder amount”, “heating temperature—fillet area”, “temperature—solder viscosity”, “solder viscosity—solder amount”, “solder amount—fillet area”. "" Is set to "". Then, as shown in FIG. 16B, these variables are connected by solid arrows.

  Here, FIG. 17 is an explanatory diagram illustrating an example of a display method by the causal structure output editing unit 14.

  The causal structure output editing unit 14 has a function of displaying only the route of only the target variable of interest. As shown in FIG. 17, when the amount of solder is an objective variable, only the route to the amount of solder is displayed. Thus, displaying only the route to the target variable of interest makes it easier to see the complex causal structure. Note that a route including a variable of interest, that is, not only a variable upstream of the variable of interest but also a downstream variable may be displayed. Further, only variables downstream from the variable of interest may be displayed.

  The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope shown in the claims. That is, embodiments obtained by combining technical means appropriately modified within the scope of the claims are also included in the technical scope of the present invention.

  The present invention can be configured as follows, for example.

  The causal structure acquisition method of the present invention is a method for determining whether or not there is a causal relationship between variables, not by performing a test for independence with a fixed threshold value, but by selecting a test with a large test statistic. It may be. For example, a test statistic with a standard deviation of 1σ or more is selected.

  Further, the causal structure acquisition method of the present invention may be a method of setting an arrow line until a route to the target variable of interest is found when there is no route.

  In addition, the causal structure acquisition method of the present invention may be a method of classifying variables into types and deleting unnecessary arrow lines in accordance with their properties.

  In addition, the causal structure acquisition method of the present invention deletes or adds an arrow line by using information input as prior knowledge such as the order of variables, types of variables, or presence / absence of arrow lines in advance. It may be a method.

  In addition, the causal structure acquisition method of the present invention may be a method of setting the preceding relationship of variables and the type of variables at the same time.

  In addition, the causal structure acquisition method of the present invention may be a method of sorting relationships between variables in the order of prior relationship. Thereby, a contradiction can be easily determined.

  Further, the causal structure acquisition method of the present invention may be a method of displaying only the route to the target variable of interest. Further, the causal structure may be displayed on the screen, and the line type (solid line, dotted line, line thickness, etc.) may be changed depending on the magnitude of the test statistic.

  Finally, each block of the causal structure acquisition apparatus 10, in particular, the inter-variable relationship generation unit 11, the arrow line determination unit 12, the route presence / absence determination unit 13, and the causal structure output editing unit 14 may be configured by hardware logic. Alternatively, it may be realized by software using a CPU as follows.

  That is, the causal structure acquisition apparatus 10 includes a central processing unit (CPU) that executes instructions of a control program that realizes each function, a read only memory (ROM) that stores the program, and a random access memory (RAM) that expands the program. ), A storage device (recording medium) such as a memory for storing the program and various data. An object of the present invention is to read a program code (execution format program, intermediate code program, source program) of a control program (causal structure acquisition program) of the causal structure acquisition apparatus 10 which is software that realizes the functions described above by a computer. This can also be achieved by supplying the recording medium recorded as possible to the causal structure acquisition apparatus 10 and reading and executing the program code recorded on the recording medium by the computer (or CPU or MPU).

  Examples of the recording medium include a tape system such as a magnetic tape and a cassette tape, a magnetic disk such as a floppy (registered trademark) disk / hard disk, and an optical disk such as a CD-ROM / MO / MD / DVD / CD-R. Card system such as IC card, IC card (including memory card) / optical card, or semiconductor memory system such as mask ROM / EPROM / EEPROM / flash ROM.

  Further, the causal structure acquisition apparatus 10 may be configured to be connectable to a communication network, and the program code may be supplied via the communication network. The communication network is not particularly limited. For example, the Internet, intranet, extranet, LAN, ISDN, VAN, CATV communication network, virtual private network, telephone line network, mobile communication network, satellite communication. A net or the like is available. Also, the transmission medium constituting the communication network is not particularly limited. For example, even in the case of wired such as IEEE 1394, USB, power line carrier, cable TV line, telephone line, ADSL line, etc., infrared rays such as IrDA and remote control, Bluetooth ( (Registered trademark), 802.11 wireless, HDR, mobile phone network, satellite line, terrestrial digital network, and the like can also be used. The present invention can also be realized in the form of a computer data signal embedded in a carrier wave in which the program code is embodied by electronic transmission.

  Since the causal structure of knowledge information can be easily obtained, it can be widely applied to the construction of inference systems such as diagnosis and prediction using the causal structure.

It is a functional block diagram which shows the outline of a structure of the causal structure acquisition apparatus which concerns on one embodiment of this invention. It is explanatory drawing which shows the outline of the function of the causal structure acquisition apparatus shown in FIG. It is explanatory drawing which shows the specific example of the table of the item name of the variable stored in the prior relationship storage part of the causal structure acquisition apparatus shown in FIG. It is a flowchart which shows the outline of the process in the causal structure acquisition apparatus shown in FIG. It is explanatory drawing of the kind of variable handled with the causal structure acquisition apparatus shown in FIG. In the causal structure acquisition apparatus shown in FIG. 1, it is a screen display example for a user to input a variable and a classification of a variable. It is a flowchart which shows the detail of the process which determines the causal structure from data in the causal structure acquisition apparatus shown in FIG. It is explanatory drawing explaining the method in which the variable relationship production | generation part of the causal structure acquisition apparatus shown in FIG. 1 determines the grade of the causal relationship between variables from a test statistic. It is explanatory drawing of the specific example which concerns on embodiment, (a) is a relationship table between variables, (b) is a causal structure displayed based on the relationship table between variables of (a). It is explanatory drawing of the specific example which concerns on embodiment, (a) is a relationship table between variables, (b) is a causal structure displayed based on the relationship table between variables of (a). It is explanatory drawing which shows the process which deletes an unnecessary arrow line automatically using the property by the kind of variable in the causal structure acquisition apparatus shown in FIG. It is explanatory drawing of the specific example which concerns on embodiment, and shows the relationship table between variables. It is explanatory drawing of the specific example which concerns on embodiment, and shows the relationship table between variables. It is explanatory drawing of the specific example which concerns on embodiment, and shows the relationship table between variables. It is a flowchart which shows the detail of a process of the causal structure acquisition apparatus shown in FIG. It is explanatory drawing of the specific example which concerns on embodiment, (a) is a relationship table between variables, (b) is a causal structure displayed based on the relationship table between variables of (a). It is explanatory drawing which shows an example of the display system by the causal structure output edit part of the causal structure acquisition apparatus shown in FIG.

Explanation of symbols

10 causal structure acquisition device 11 inter-variable relationship generating unit (inter-variable relationship generating means)
12 arrow line determination unit 13 route presence / absence determination unit 14 causal structure output editing unit (display means)
15 Data Input Unit 21 Variable Storage Unit 22 for Each Type Preceding Relationship Storage Unit 23 Measurement Data Storage Unit

Claims (9)

A plurality of sets of measurement data of two variables selected from a variable group including a plurality of variables are acquired from the measurement data storage unit, and an independence test statistic of the two variables is obtained based on the acquired measurement data. If the test statistic is larger than the upper threshold, there is a causal relationship between the two variables, and if the test statistic is smaller than the lower threshold, there is no causal relationship between the two variables. If the quantity is between two thresholds, the variable relation generating means for determining that the two states are in a first state where it is not known whether or not there is a causal relationship between the two variables ,
The inter-variable relationship generating means further sets a variation threshold smaller than the lower threshold, and when the test statistic is between the variation threshold and the lower threshold, there is a causal relationship between the two variables. Whether or not there is a causal relationship between the two variables of interest, directly or indirectly through other variables, while determining that the second state is unknown. A causal structure acquisition apparatus characterized in that the fluctuation threshold value is reduced until it is determined that the state is unknown . The inter-variable relationship generation means changes the presence / absence of the determined causal relationship according to causal relationship definition information that predefines the presence / absence of a causal relationship with another variable for each variable. Causal structure acquisition device. The inter-variable relationship generating means assigns the element to each element of a two-dimensional matrix in which variables located upstream and downstream are arranged in rows and columns in a route connecting variables having causal relationships. A variable relation table in which presence / absence of a causal relationship between two specified variables is set is generated, and the row and column variables are rearranged in the same order and in a preceding relation with respect to the rows and columns. The causal structure acquisition apparatus according to 1 or 2. A connection line that connects the two variables in the first state, which does not know whether or not there is a causal relationship, and a connection line that connects the two variables in the second state that do not know whether or not there is a causal relationship The causal structure acquisition apparatus according to any one of claims 1 to 3, further comprising display means for displaying with seeds. It said display means, the variables included in the variable group, the causal structure acquisition apparatus of claim 4, wherein the benzalkonium be displayed together with the connection line for connecting the causal relationship two variables.   The display means displays only a variable of interest and a variable causally related to the variable of interest directly or indirectly through another variable together with the connection line. Item 6. The causal structure acquisition device according to Item 5. A causal structure acquisition method by a causal structure acquisition apparatus,
The inter-variable relationship generation means provided in the causal structure acquisition apparatus acquires a plurality of sets of measurement data of two variables selected from a variable group including a plurality of variables from the measurement data storage unit, and based on the acquired measurement data If the test statistic is greater than the upper threshold, there is a causal relationship between the two variables, and if the test statistic is smaller than the lower threshold, There is no causal relationship between two variables. If the test statistic is between two thresholds, the relationship between variables is determined to be a first state in which it is not known whether there is a causal relationship between the two variables. a generation step seen including,
In the inter-variable relationship generation step, a variation threshold smaller than the lower threshold is set, and when the test statistic is between the variation threshold and the lower threshold, there is a causal relationship between the two variables. Whether or not there is a causal relationship between the two variables of interest, directly or indirectly through other variables, while determining that the second state is unknown. A causal structure acquisition method characterized by reducing the fluctuation threshold until it is determined that the state is unknown .   A causal structure acquisition program for realizing the causal structure acquisition apparatus according to any one of claims 1 to 6 by a computer and causing the computer to function as each means.   A computer-readable recording medium on which the causal structure acquisition program according to claim 8 is recorded.

Images