JPS60157640A - Prolog execution controlling system - Google Patents

Abstract

PURPOSE:To quicken a re-output processing time, and to control dealing of a result on the way by accumulating plural answers, and postponing the processing, in a program processing system using a predicate logical type language program. CONSTITUTION:Basing on a question whose processing is requested from a terminal equipment 5, an execution control part 2 generates a goal list, starts a process in order to solve it, and delivers a new goal list to its process. In case a body of its goal list is constituted of plural literals C, D, in case there are plural D, -G answers of the goal list, only a part of the answers is returned to a high-order process, and the rest is accumulated in a memory 4 and its processing is postponed. In case a re-execution is requested from the high-order process, other accumulated answers are outputted successively.

Description

[Detailed Description of the Invention] [Technical Field of the Invention] The present invention relates to an efficient execution control method in a program processing system using the predicate logic language PROLOG, and in the process of parallel processing of PROLOG programs to 2%. If multiple intermediate results occur, by keeping them intermediate.

The present invention relates to an execution control method that suppresses future processing, temporarily postpones processing that is unnecessary at the present time, and then pressures it to respond when the need arises.

[Technology background]

A knowledge-paced processing system using the Pl'tOLOG program finds rules that match the question.

Next, the process of finding facts (constants) that match the variables included in the rule and obtaining the answer is performed. When served.

There are many combinations of facts that satisfy the rule.

The amount of logical processing and data transfer increases significantly.

However, only a part of the process actually leads to the correct answer, and there are many intermediate processes that end up being wasted.
This cannot be known in advance during the processing process.

On the other hand, PROLOG programs can be made more efficient through parallel processing, and several types of execution control methods have been proposed. Examples include the PIE model, reduction type model, C0NBRY model, and MP type model.

Among these, the PIE type model and the Dakushi type model are methods in which when there are multiple intermediate results, they are all processed without being retained in the middle, so if another answer is requested, it can be easily obtained. . but.

The former has the disadvantage that the amount of data transferred increases because the processing units are made completely independent, and the amount of required resources increases because all the processing units that can be executed on both sides are processed. On the other hand, in the case of the C0NE Y model and the MP model, multiple intermediate results occur.

OR process and 5EN, respectively, except for some results
It is stored in a process with a specific function called the DER process, and further processing is put on hold.

If a different answer is later requested, the new answer is determined using the accumulated intermediate results, thereby improving processing efficiency. However, in that case, in order to obtain a new answer, the value assigned in the previous answer is replaced with another value (intermediate result).
When redoing the process by replacing it with
[Objective and Structure of the Invention] The object of the present invention is to efficiently solve the problem with a small processing time and resource amount when there are multiple intermediate results under the program processing pressure of PROLOG. The objective is to provide a means for re-executing (RIEDo) the program, and its configuration consists of a knowledge base storage means, a means for generating a goal list based on a question, and a means for generating a goal list based on a question in the PROLOG program processing system. and processing means provided for each generated goal list for performing unification with the clauses in the knowledge base.

When there is one literal in the body part of the goal list.

If unification generates a new goal list for a successful close and activates a new processing means to solve the goal list, while there are multiple literals in the body part of the goal list.

Unification is performed on one of the literals, and if the unification is successful, a new degenerate goal list is generated by assigning the body of the successful close to the literal, and a new process is performed to solve the goal list. When multiple answers are obtained in the processing means that activates the means and solves any of the goal lists.

The present invention is characterized in that the plurality of answers are accumulated in the processing means for solving the non-degenerate goal list, and the processing is carried out later by excluding some of the plurality of answers.

[Embodiments of the invention]

The details of the present invention will be explained below based on two examples.

FIG. 1 is a functional configuration diagram of a PROLOG processing system according to an embodiment of the present invention method, where l is a PROLOG processing device, 2 is an execution control unit, 3 is a process started for each goal list, and 4 is a knowledge base. A storage device and 5 indicate a terminal device.

FIG. 2 shows an example of PROLOG program processing 311 for explaining the functions of the embodiment system shown in FIG.
In order to obtain an answer to the question ``'i'A Knee B'' using the given knowledge base, the goal list that is sequentially generated and the processes ■ to θ etc. activated for the processing are shown. It is something that

A single solid line arrow indicates a forward processing route where the variable value is entered, and a double solid line arrow indicates a backward processing route where the found value is output as the answer. Process with diagonal lines ■l o
+ θ is a non-degenerate process that accumulates once multiple answers come up from lower processes (
(described later).

The dotted arrow indicates a re-execution (WDO) route that requests the output of other answers stored in the non-degenerate process if the answer currently being output by fA is not satisfied.

In FIG. 1, the line control unit 2 generates a goal list based on a question requested for processing by the terminal device 5, and starts a process for unifying the goal list with the knowledge base. The process can further activate other processes according to the processing results.

The basic function of the process is as follows.

(1) Receive the goal list and perform secondary processing.

■ When the body of the goal list consists of one literal (for example, A knee B in Figure 2).

0 Unify the literal with the knowledge base, generate a new goal list for a successful closure (rule or fact), launch a process to solve it, and add the new goal list to that process. hand over. Generally, there are multiple successful closes, so a new goal list is created for each one (starting the process; that is, branching is performed).

■ When the body of the goal list is composed of multiple literals (for example, B!-C, D in FIG. 2).

0 Performs unification on one of multiple literals. Generate a new goal list by filling it in place of the corresponding literal in the goal list that is the body of a successful closing (called goal list degeneracy), start a process to solve it, and add that new goal to that process. Pass the list.

(2) The answers to the goal list are stored only in the process of solving the first non-reduced goal list.

If there are multiple answers, only some of the answers are returned to the upper process, the rest are accumulated, and their processing is postponed.

For example, in FIG. 2, process θ has multiple answers to D knee G, but only the first one of them is sent to process e.

(3) The answer from each goal list is basically Niha.

It propagates only between processes that hold the answer to (2) (double solid line arrow and dotted line arrow in Figure 2).

Therefore, in the example shown in Fig. 2, the body has multiple lites 9/L.
/ Omotsu goal list, for example, a goal list B and C with two literals C and D connected by AND logic.
, D, when multiple answers are obtained in the process θ, [phase], ■, which processes the sequentially degenerated goal list, the process @ that processes The respective answers are not stored in θ, ■, and ■.

The same holds true for other non-degenerate processes ① and θ. That is, multiple answers in processing a degenerate goal list should be concentrated on the process of the first goal list that is not degenerated, and then the process for refilling.
Back tracking to O2@
) is unnecessary.

As a result, when there are multiple intermediate results, all of them can be processed in advance, reducing the processing burden.
However, when re-execution (REDO) is requested from the upper process, only processes ■, @, and θ need to sequentially output the other answers they have stored, which speeds up the re-output process. can do.

The following knowledge base and questions are examples.

The process of executing a program according to the method of the present invention will be described in detail.

Knowledge base (Dl) Father (Abe, Morio) / *Abe's father is Morio */ (D2) Father (Morio, Taira) / * Morio's father is Taira +/ (D3) Father (Uko 2 Kenji) / *Yoko's father is Chapter 2 * / (D4) Mother (Tabe, Yoko) / *Tabe's mother is
Yoko */(D5) grandfather CX, gt)) knee father (x
*par)s father (pare gp)/*If x's father's father is gp, then 2g-sho is X's grandfather*〆(D
6) Grandfather Cx*gp>nee Mother It t'ar)*father<p
ar, ip>/*If x's mother's father is gp, then 9gp
Is X's grandfather *〆Question (q)? Grandfather (Tabe, gp) Program execution process Here, we will explain the program execution process (11 (Process until the value of gp in (G) is determined (forward processing)) (2) The process until the answer to (G) is output. 11 (backward processing) This will be explained in 02 stages.

(Process until the value of gp in 11(G) is determined (forward processing): See FIG. 3.

1)・? Given the grandfather (Tabe, ip>, the output (
1p)', create a grandfather (Abe, 111).

・Create a process.

・Output (gp) Requests process ■ to process the output (gp).

2) Succeeded with knowledge base (D5) and (D6) with @Grandfather's (Abe, gtJ) UNIFICE gastrointestinal.

・Generate processes ■ and ■.

・D5. Set D6 as the new goal list [, respectively ■, ■
request processing.

3)・■ My father's (Abe + par) Unifi-Kashi adventure (Dl) and success c part Morio).

mPar = Morio is reflected in the goal list, grandfather (Abe, i
p>: Father (Abe, 9 husbands), Father (9 husbands.

gp).

Treat yourself condescendingly. Grandfather (Tabe, IP), father (9 husbands).

gp).

Force＼ Since there are literals to be further processed, process■
Generate a.

ψGrandfather (Tabe, GP) - Father (9th husband, GP) is requested to be processed by ■ as a new goal list.

In case (2), the same processing as in (2) is performed.

4)・Father at @ (9 husbands, GP>'s uniform!
D2) and successful Cgp-hei).

・GP=Taira is reflected in the goal list. Grandfather (Abe, Taira), father (Morio, Taira).

- There are no more literals to process. That's why I got the answer.

(Process until the answer to 21(G) is output (backward processing): See FIG. 4.

5)・■ is the grandfather of the head section of the goal list (Father).

Send (Taira) to ■.

・■ is the grandfather of the head section of the goal list (Father).

Send Chapter 2) to ■.

6)・■ sends the grandfather (Abe, Taira) to ■ as the answer to the goal list.

・■ is my grandfather (Tabe 9, chapter 2) as the answer to the goal list■
send to

For example, if multiple answers are returned to ■, ■, etc., some of the answers can be sent to ■, and the rest can be retained as intermediate results and processed later.

7) Reflect the value of the argument of the grandfather (Abe, 'F) received from the battle ■ in the goal list. Output (Taira) Nee Grandfather (Tabe, Taira)
.

・Same as ■ for ■. Output (Chapter 2); - Grandfather (Tabe 1 Chapter 2) - Output (Taira) as the answer. Output (chapter 2) can be output or held using ■.

〔Effect of the invention〕

As described above, according to the second aspect of the present invention, the process that is accumulated when there are multiple intermediate results is limited to the process that solves the goal list that is not degenerate, so efficient re-execution processing without backtracking is possible. This makes it possible to speed up the re-output processing time compared to the conventional method, and also makes it possible to handle and control intermediate results.

[Brief explanation of drawings]

FIG. 1 is a configuration diagram of a system according to an embodiment of the present invention. FIG. 2 is an explanatory diagram of a processing example in the embodiment of the present invention. FIG. 3 is an explanatory diagram of a specific example of forward processing. FIG. 4 is an explanatory diagram of a specific example of backward processing. In the figure, 1 is a PROLOG processing device, 2 is an execution control unit, and 3
is a process, 4 is a knowledge base storage device, and 5 is a terminal device. Patent applicant Fujitsu Ltd. Representative patent attorney Fumihiro Hase (1 other person)

Claims (1)

[Claims] In a PROLOG program processing system. comprising knowledge base storage means, means for generating a goal list based on a question, and processing means provided for each generated goal list for unifying the generated goal list with closings in the knowledge base; If the body part of the goal list has one literal, the processing means generates a new goal list for a close in which unification is successful, and activates a new processing means to solve the goal list. , on the other hand, if there are multiple literals in the body part of the goal list. Unification is performed on one of the literals, and if the unification is successful, a new degenerate goal list is generated by assigning the body of the successful close to the literal, and a new process is performed to solve the goal list. When multiple answers are obtained in the processing means that activates the means and solves any of the goal lists. A PROLO characterized in that the plurality of answers are accumulated in the processing means for solving the non-degenerate goal list, and processing is carried out later by excluding some of the plurality of answers.
G execution control method.