New unit tests for NAL 5 & 6 ( work in progress )

[Jean-Marc Vanel]

Hi

I added in new directory nars-dist/tests , unit tests for NAL 5 and 6 .

http://code.google.com/p/open-nars/source/browse/#svn%2Ftrunk%2Fnars-dist%2Ftests%2Fnal5

http://code.google.com/p/open-nars/source/browse/#svn%2Ftrunk%2Fnars-dist%2Ftests%2Fnal6

The results looks OK for  NAL 5 .

But for NAL 6, only the one from  "Table 8.3: Sample Independent-Variable Introduction Rules" works :

**** Induction 

<M --> P>.

<M --> S>.

<<$x --> S> ==> <$x --> P>>?

4

*

The ones I made by combining Table 8.3 and Table 3.3 ( Basic Syllogistic Rules ) do not bear any result , like :

**** Exemplification , Deduction : no result !!!

< P --> M >.

< M --> S >.

<<$x --> S> ==> <$x --> P>>?

<<$x --> P> ==> <$x --> S>>?

600

The tests in variable-elimination-in.nars also do not bear any result , like :

// Table 8.2: Sample Independent-Variable Elimination Rules

// ded

<<#x --> S> ==> <#x --> P >>.

< M --> S >.

<M --> P >?

600

--
Jean-Marc Vanel
Déductions SARL - Consulting, services, training,
Rule-based programming, Semantic Web
http://deductions-software.com/
+33 (0)6 89 16 29 52
Twitter: @jmvanel ; chat: irc://irc.freenode.net#eulergui

[Pei Wang]

Both of your examples should not produce result in NARS. See http://www.cis.temple.edu/~pwang/Writing/NAL-Specification.pdf , page 50:

"an independent variable
normally appears in both sides of an implication or equivalence copula, as
extension or intension of two terms. A dependent variable normally appear
in two components of a conjunction, also as extension or intension of two
terms."

The valid inference patterns of NAL-6 are listed in http://code.google.com/p/open-nars/source/browse/trunk/nars-dist/Examples/Example-NAL6-in.txt which should be the basis of unit test.

Pei

--
You received this message because you are subscribed to the Google Groups "open-nars" group.
To unsubscribe from this group and stop receiving emails from it, send an email to open-nars+...@googlegroups.com.
To post to this group, send email to open...@googlegroups.com.
Visit this group at http://groups.google.com/group/open-nars?hl=en.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

[Jean-Marc Vanel]

2013/4/2 Pei Wang <mail.p...@gmail.com>

Both of your examples should not produce result in NARS. See http://www.cis.temple.edu/~pwang/Writing/NAL-Specification.pdf , page 50:

"an independent variable
normally appears in both sides of an implication or equivalence copula, as
extension or intension of two terms. A dependent variable normally appear
in two components of a conjunction, also as extension or intension of two
terms."

 

I'm sorry, but both the working example and the non working example both satisfy this.

To create the tests , I tried to interpret what NAL-Specification.pdf says, page 52 :

"Table 8.3: Sample Independent-Variable Introduction Rules"

The valid inference patterns of NAL-6 are listed in http://code.google.com/p/open-nars/source/browse/trunk/nars-dist/Examples/Example-NAL6-in.txt which should be the basis of unit test.

Pei

On Tue, Apr 2, 2013 at 5:03 AM, Jean-Marc Vanel <jeanmar...@gmail.com> wrote:

Hi

I added in new directory nars-dist/tests , unit tests for NAL 5 and 6 .

http://code.google.com/p/open-nars/source/browse/#svn%2Ftrunk%2Fnars-dist%2Ftests%2Fnal5

http://code.google.com/p/open-nars/source/browse/#svn%2Ftrunk%2Fnars-dist%2Ftests%2Fnal6

The results looks OK for  NAL 5 .

But for NAL 6, only the one from  "Table 8.3: Sample Independent-Variable Introduction Rules" works :

**** Induction 

<M --> P>.

<M --> S>.

<<$x --> S> ==> <$x --> P>>?

4

*

The ones I made by combining Table 8.3 and Table 3.3 ( Basic Syllogistic Rules ) do not bear any result , like :

**** Exemplification , Deduction : no result !!!

< P --> M >.

< M --> S >.

<<$x --> S> ==> <$x --> P>>?

<<$x --> P> ==> <$x --> S>>?

600

I didn't realize that the scope of $x spans over both lines. 

Decomposed in two runs, this still does not work :

< P --> M >.

< M --> S >.

<<$x --> S> ==> <$x --> P>>?

6

***

< P --> M >.

< M --> S >.

<<$y --> P> ==> <$y --> S>>?

6

And in one run, this still does not work :

< P --> M >.

< M --> S >.

<<$x --> S> ==> <$x --> P>>?

<<$y --> P> ==> <$y --> S>>?

6

And it is understood by NARS as this :

**** Deduction , Exemplification : no result !!!

  IN: <M --> P>. %1.00;0.90% {0 : 1} 

  IN: <S --> M>. %1.00;0.90% {0 : 2} 

  IN: <<$1 --> S> ==> <$1 --> P>>?  {0 : 3} 

  IN: <<$1 --> P> ==> <$1 --> S>>?  {0 : 4} 

Both $x and $y are translated into $1 .

So (side question ) it appears to be impossible to ask 2 questions involving 2 different variables.

[Jean-Marc Vanel]

I had a closer look at Example-NAL6-in.txt ,

and now I understand what ' s wrong with my tests.

I'll try to leverage on that to add the missing cases that must work :) .

2013/4/2 Pei Wang <mail.p...@gmail.com>

[Jean-Marc Vanel]

I read 2013 book and NAL-specification.pdf, but found it difficult to infer from the text a complete set of unit elementary inferences.

Several key tables, like Table 8.3 in NAL-specification.pdf , are labeled "sample rules" , but I did not understand how to derive all the rules, and a few trials I made were unsuccessful.

Not related, I updated issue 24 :

http://code.google.com/p/open-nars/issues/detail?id=24 .

2013/4/2 Jean-Marc Vanel <jeanmar...@gmail.com>

[Pei Wang]

To list every rule will make the writing too tedious to read, though I did make the effort to summarize them in the appendices.

You can also see the example files as a list of rules, since all major variations are given in them.

Pei
 

[Pei Wang]

Or you can see the Prolog version of NAL in http://www.cis.temple.edu/~pwang/Implementation/NAL/nal.pl , which can be seen as a complete list of inference rules from NAL-1 to NAL-6.

Pei

[Jean-Marc Vanel]

Indeed, the inference rules are listed in the appendices, but not the rules for variables introduction and elimination.

I guess if I read more carefully chapters on NARS6 I'll understand.

I guess  the rules for variables are not possibly enumerated, because it they are introduced by unification.

About varaible introduction, I'm still try to corner this regression with respect to NARS 1.3.3 , cf 

http://code.google.com/p/open-nars/issues/detail?id=24

<b1 --> blue>.
<(*,r1,b1) --> pattern1>.
<<$1 --> blue> ==> <(*,r1,$1) --> pattern1>>?

Fails with 1.5.1 , succeeds in 263 steps in 1.3.3 .

2013/4/4 Pei Wang <mail.p...@gmail.com>

[Jean-Marc Vanel]

About variable introduction, I wrote a unit test that fails :

https://open-nars.googlecode.com/svn/trunk/nars-dist/tests/nal6/multi-variable_introduction-in.nars

and is a direct translation of "PDF" dialect into Open-NARS dialect.

Also, there are many lines commented out in class CompositionalRules ,

in methods 

introVarOuter() and introVarInner() .

What is the status of these commented lines ?

Future features ?

Code in need of reworking?

Could it be that the regressions are caused by this code commenting ?

2013/4/4 Jean-Marc Vanel <jeanmar...@gmail.com>

[Pei Wang]

On Thu, Apr 4, 2013 at 10:38 AM, Jean-Marc Vanel <jeanmar...@gmail.com> wrote:

About variable introduction, I wrote a unit test that fails :

https://open-nars.googlecode.com/svn/trunk/nars-dist/tests/nal6/multi-variable_introduction-in.nars

and is a direct translation of "PDF" dialect into Open-NARS dialect.

Replace the `#' by `$'.

 

Also, there are many lines commented out in class CompositionalRules ,

in methods 

introVarOuter() and introVarInner() .

What is the status of these commented lines ?

Future features ?

Code in need of reworking?

When 1.5.0 came from 1.4.0, many lines are commented out. Some are the code for NAL-7 and above, and some are caused by tentative design changes. I have to check them one by one to decide the reason and status.

 

Could it be that the regressions are caused by this code commenting ?

That is possible, though I won't have any time to touch the code before April 20, so cannot say for sure.

Pei
 

[Jean-Marc Vanel]

2013/4/4 Pei Wang <mail.p...@gmail.com>

On Thu, Apr 4, 2013 at 10:38 AM, Jean-Marc Vanel <jeanmar...@gmail.com> wrote:

About variable introduction, I wrote a unit test that fails :

https://open-nars.googlecode.com/svn/trunk/nars-dist/tests/nal6/multi-variable_introduction-in.nars

and is a direct translation of "PDF" dialect into Open-NARS dialect.

Replace the `#' by `$'.

I know that of course :) .

I expected some comment on the fact that this rule does not fire in currrent code ( follow link above ).

 

Also, there are many lines commented out in class CompositionalRules ,

in methods 

introVarOuter() and introVarInner() .

What is the status of these commented lines ?

Future features ?

Code in need of reworking?

When 1.5.0 came from 1.4.0, many lines are commented out. Some are the code for NAL-7 and above, and some are caused by tentative design changes. I have to check them one by one to decide the reason and status.

You probably should have forked from the revision 1.3.3 release to create 1.5 .

And / or the 1.4 should have been a separate branch on the depot.

But that 's the past, let's look at future !

 

Could it be that the regressions are caused by this code commenting ?

That is possible, though I won't have any time to touch the code before April 20, so cannot say for sure.

Do you see any other path I could check before April 20 ?

Does it help to have this more simple regression here (look at end of page ) :

http://code.google.com/p/open-nars/issues/detail?id=24

[Pei Wang]

On Thu, Apr 4, 2013 at 3:27 PM, Jean-Marc Vanel <jeanmar...@gmail.com> wrote:

2013/4/4 Pei Wang <mail.p...@gmail.com>

On Thu, Apr 4, 2013 at 10:38 AM, Jean-Marc Vanel <jeanmar...@gmail.com> wrote:

About variable introduction, I wrote a unit test that fails :

https://open-nars.googlecode.com/svn/trunk/nars-dist/tests/nal6/multi-variable_introduction-in.nars

and is a direct translation of "PDF" dialect into Open-NARS dialect.

Replace the `#' by `$'.

I know that of course :) .

I expected some comment on the fact that this rule does not fire in currrent code ( follow link above ).

I'll have to trace it to make that comment. ;-)
 

 

Also, there are many lines commented out in class CompositionalRules ,

in methods 

introVarOuter() and introVarInner() .

What is the status of these commented lines ?

Future features ?

Code in need of reworking?

When 1.5.0 came from 1.4.0, many lines are commented out. Some are the code for NAL-7 and above, and some are caused by tentative design changes. I have to check them one by one to decide the reason and status.

You probably should have forked from the revision 1.3.3 release to create 1.5 .

And / or the 1.4 should have been a separate branch on the depot.

But that 's the past, let's look at future !

1.5 also use 1.4 code that is not in 1.3.3. Also, when I coded 1.4, I didn't expect to go back to 1.3.3.

 

 

Could it be that the regressions are caused by this code commenting ?

That is possible, though I won't have any time to touch the code before April 20, so cannot say for sure.

Do you see any other path I could check before April 20 ?

No --- sorry.

Does it help to have this more simple regression here (look at end of page ) :

http://code.google.com/p/open-nars/issues/detail?id=24

Probably,  but it may be easier to start from NAL-6 example

********** multiple variables introduction
  IN: <{key1} --> (/,open,_,{lock1})>.
  IN: <{key1} --> key>.
2
 OUT: <<$1 --> key> ==> <$1 --> (/,open,_,{lock1})>>. %1.00;0.45% {2 : 1;2}

It works in 1.5.2, and is basically the same problem.

Pei

[Jean-Marc Vanel]

"my" regression's question can be rewritten:

<<#1 --> blue>==>< #1 --> (/,pattern1,r1,_)> ?

So I tried this:

<b1 --> blue>.
<(*,r1,b1

) --> pattern1>.
<<$1 --> blue> ==> < $1 --> (/,pattern1,r1,_)> ?
100

This works ( with silence == 62 ) :

46
 OUT: <b1 --> blue>?  {46 : 2;4} 
26
 OUT: <<$1 --> blue> ==> <$1 --> (/,pattern1,r1,_)>>. %1.00;0.45% {71 : 1;2} 

Then the last step to desired conclusion is :

<<$1 --> blue> ==> <$1 --> (/,pattern1,r1,_)>>.
<<$1 --> blue> ==> <(*,r1,$1) --> pattern1>>?
20

with result:

 OUT: <<$1 --> blue> ==> <(*,r1,$1) --> pattern1>>. %1.00;0.90% {19 : 1} 

So, I'm puzzled, because I have 2 inferences runs : A |- B , B |- C but the direct inference A |- C fails ! To be sure, I ran to 50 000 steps the direct inference :

<b1 --> blue>.
<(*,r1,b1) --> pattern1>.
<<$1 --> blue> ==> <(*,r1,$1) --> pattern1>>?

My understanding was that to debug a NARS case, each step could separately checked, and the global direct inference should then work.

[Pei Wang]

It needs to be traced. In general your understanding is correct, though due to resource restriction, not all possible inference will be realized. However, for such a simple problem, it should.

If you cannot resolve the issue soon, add it to the issue list for me to handle later.

Pei

[Jean-Marc Vanel]

All the story is in issue 24 :

http://code.google.com/p/open-nars/issues/detail?id=24

Indeed, I'm out of ideas to check.

When you write "It needs to be traced" , do you mean, examining the log window, comparing the successful run with the unsuccessful, and looking for specific concepts activation and  "!!! Derived:" messages ?

2013/4/5 Pei Wang <mail.p...@gmail.com>

[Jean-Marc Vanel]

Assuming the answer is yes to preceding question, I ran again the direct (failing) inference :

<b1 --> blue>.
<(*,r1,b1) --> pattern1>.
<<$1 --> blue> ==> <(*,r1,$1) --> pattern1>>?

In log window, I see that this is infered at step 35 :

 from task: $0.80;0.80;0.95$ <b1 --> blue>. %1.00;0.90%  
 from belief: <b1 --> (/,pattern1,r1,_)>. %1.00;0.90% {8 : 2} 
!!! Derived: $0.1718;0.1270;0.2230$ <<$1 --> blue> ==> <$1 --> (/,pattern1,r1,_)>>. %1.00;0.45%  {35 : 2;1}   

It is re-infered several times later, but the intentional image is never translated to the desired pattern.

2013/4/5 Jean-Marc Vanel <jeanmar...@gmail.com>

[Pei Wang]

Then it is the code that has to be traced to see why the corresponding rule is not triggered.

Pei

[Pei Wang]

Translation works fine:

  IN: <<$1 --> blue> ==> <$1 --> (/,pattern1,r1,_)>>. %1.00;0.90% {0 : 1}
3
 OUT: <<$1 --> blue> ==> <r1 --> (/,pattern1,_,$1)>>. %1.00;0.90% {3 : 1}
3
 OUT: <<$1 --> blue> ==> <(*,r1,$1) --> pattern1>>. %1.00;0.90% {6 : 1}

[Jean-Marc Vanel]

As I wrote , I checked this .

By "code that has to be traced" you mean debug , but where to put breakpoint  or print ?

[Pei Wang]

Well, that will be a long story --- you need to know which rule is responsible, and where it should be triggered.

I did some experiment, and believe the rules are all fine, though the control mechanism somehow ignores that possible combination --- maybe it got a low priority value, then it is forgot. You'll have to wait until I have more time for it. Sorry again.

Pei

[Jean-Marc Vanel]

Do not worry ! Even your short messages help .

Currently I'm deciphering the NARS 1.3.3 log that succeeds in 263 steps ...

2013/4/5 Pei Wang <mail.p...@gmail.com>

[Jean-Marc Vanel]

Deciphering the NARS 1.3.3 log that succeeds in 263 steps ...

Quite early a form very near the answer is obtained:

--- 19 ---
 * Selected Concept: b1
 * Selected TaskLink: $0.9426;0.8000;0.9487$  _@(T4-1) <b1 --> blue>. %1.0000;0.9000% {0 : 2} 
 * Selected TermLink: $0.1421;0.3394;0.4243$  _@(T4-1) <b1 --> (/,pattern1,r1,_)>
 * Selected Belief: <b1 --> (/,pattern1,r1,_)>. %1.0000;0.9000% {10 : 3} 
!!! Derived: $0.3628;0.3811;0.6690$ <(/,pattern1,r1,_) --> blue>. %1.0000;0.4475% {19 : 3;2} 
!!! Derived: $0.5548;0.3811;0.6690$ <blue --> (/,pattern1,r1,_)>. %1.0000;0.4475% {19 : 3;2} 
!!! Derived: $0.6183;0.3811;0.6690$ <(/,pattern1,r1,_) <-> blue>. %1.0000;0.4475% {19 : 3;2} 
!!! Derived: $0.3628;0.3811;0.6690$ <<#1 --> (/,pattern1,r1,_)>==><#1 --> blue>>. %1.0000;0.4475% {19 : 3;2} 
!!! Derived: $0.5548;0.3811;0.6690$ <<#1 --> blue>==><#1 --> (/,pattern1,r1,_)>>. %1.0000;0.4475% {19 : 3;2}

It is re-infered several times later.

But only much later is it translated to the desired pattern :

--- 262 ---
 * Selected Concept: pattern1
 * Selected TaskLink: $0.3147;0.3811;0.6690$  _@(T8-2-2-2) <<#1 --> blue> ==> <#1 --> (/,pattern1,r1,_)>>. %1.0000;0.4475% {114 : 3;2} 
!!! Derived: $0.0702;0.0850;0.2230$ <<#1 --> blue>==><(*,r1,#1) --> pattern1>>. %1.0000;0.4475% {262 : 3;2} 

Indeed it suggests that the control  and/or  priority mechanism is faulty.

I hope it's a plain bug, and that raising some priority for this case will not prevent other inferences .

2013/4/5 Jean-Marc Vanel <jeanmar...@gmail.com>

[Pei Wang]

The current control part is very rough, and needs lots of tuning and refining for sure.

Pei

[Jean-Marc Vanel]

Please read below ...

2013/4/5 Jean-Marc Vanel <jeanmar...@gmail.com>

Deciphering the NARS 1.3.3 log that succeeds in 263 steps ...

Quite early a form very near the answer is obtained:

--- 19 ---
 * Selected Concept: b1
 * Selected TaskLink: $0.9426;0.8000;0.9487$  _@(T4-1) <b1 --> blue>. %1.0000;0.9000% {0 : 2} 
 * Selected TermLink: $0.1421;0.3394;0.4243$  _@(T4-1) <b1 --> (/,pattern1,r1,_)>
 * Selected Belief: <b1 --> (/,pattern1,r1,_)>. %1.0000;0.9000% {10 : 3} 
!!! Derived: $0.3628;0.3811;0.6690$ <(/,pattern1,r1,_) --> blue>. %1.0000;0.4475% {19 : 3;2} 
!!! Derived: $0.5548;0.3811;0.6690$ <blue --> (/,pattern1,r1,_)>. %1.0000;0.4475% {19 : 3;2} 
!!! Derived: $0.6183;0.3811;0.6690$ <(/,pattern1,r1,_) <-> blue>. %1.0000;0.4475% {19 : 3;2} 
!!! Derived: $0.3628;0.3811;0.6690$ <<#1 --> (/,pattern1,r1,_)>==><#1 --> blue>>. %1.0000;0.4475% {19 : 3;2} 
!!! Derived: $0.5548;0.3811;0.6690$ <<#1 --> blue>==><#1 --> (/,pattern1,r1,_)>>. %1.0000;0.4475% {19 : 3;2}

It is re-infered several times later.

But only much later is it translated to the desired pattern :

--- 262 ---
 * Selected Concept: pattern1
 * Selected TaskLink: $0.3147;0.3811;0.6690$  _@(T8-2-2-2) <<#1 --> blue> ==> <#1 --> (/,pattern1,r1,_)>>. %1.0000;0.4475% {114 : 3;2} 
!!! Derived: $0.0702;0.0850;0.2230$ <<#1 --> blue>==><(*,r1,#1) --> pattern1>>. %1.0000;0.4475% {262 : 3;2} 

Indeed it suggests that the control  and/or  priority mechanism is faulty.

I hope it's a plain bug, and that raising some priority for this case will not prevent other inferences .

 I wonder if derivation just above at step 262 is the only way to reach the answer.

Could it be that raising this helps :

 

     /** Maximum number of goals kept in a Concept */

    public static final int MAXIMUM_QUESTIONS_LENGTH = 5;

From the NARS 1.5.2 log that fails, quite early a form very near the answer is also obtained, by the same inference as with 1.3.3 :

--- 35 ---
 * Selected Concept: b1
 * Selected TaskLink: $0.5547;0.8000;0.9487$  _@(T4-1) <b1 --> blue>. %1.00;0.90%  {0 : 1} 
 * Selected TermLink: $0.1533;0.3394;0.4243$  _@(T4-1) <b1 --> (/,pattern1,r1,_)>
 * Selected Belief: <b1 --> (/,pattern1,r1,_)>. %1.0000;0.9000% {8 : 2} 
!!! Derived: $0.2368;0.3811;0.6690$ <(/,pattern1,r1,_) --> blue>. %1.00;0.45%  {35 : 2;1}   
 from task: $0.80;0.80;0.95$ <b1 --> blue>. %1.00;0.90%  
 from belief: <b1 --> (/,pattern1,r1,_)>. %1.00;0.90% {8 : 2} 
!!! Derived: $0.4263;0.3811;0.6690$ <blue --> (/,pattern1,r1,_)>. %1.00;0.45%  {35 : 2;1}   
 from task: $0.80;0.80;0.95$ <b1 --> blue>. %1.00;0.90%  
 from belief: <b1 --> (/,pattern1,r1,_)>. %1.00;0.90% {8 : 2} 
!!! Derived: $0.4890;0.3811;0.6690$ <(/,pattern1,r1,_) <-> blue>. %1.00;0.45%  {35 : 2;1}   
 from task: $0.80;0.80;0.95$ <b1 --> blue>. %1.00;0.90%  
 from belief: <b1 --> (/,pattern1,r1,_)>. %1.00;0.90% {8 : 2} 
!!! Derived: $0.2592;0.1938;0.3402$ <b1 --> (|,(/,pattern1,r1,_),blue)>. %1.00;0.81%  {35 : 2;1}   
 from task: $0.80;0.80;0.95$ <b1 --> blue>. %1.00;0.90%  
 from belief: <b1 --> (/,pattern1,r1,_)>. %1.00;0.90% {8 : 2} 
!!! Derived: $0.2610;0.1938;0.3402$ <b1 --> (&,(/,pattern1,r1,_),blue)>. %1.00;0.81%  {35 : 2;1}   
 from task: $0.80;0.80;0.95$ <b1 --> blue>. %1.00;0.90%  
 from belief: <b1 --> (/,pattern1,r1,_)>. %1.00;0.90% {8 : 2} 
!!! Derived: $0.1718;0.1270;0.2230$ <<$1 --> blue> ==> <$1 --> (/,pattern1,r1,_)>>. %1.00;0.45%  {35 : 2;1}

but the budget value is much less. 

Could the issue lie here ?

[Pei Wang]

That may be the cause, though I need to spend much more time to study the log to be sure.

 

In general, this type of issues can be easily "fixed" by increasing the data structure size here or there, but in general it won't be a good idea to refine the control mechanism. Instead, we want the control mechanism to live with resource restrictions, even though it means to miss desired results occationally.

 

Now it is not a good time to fine tune the control mechanism, including the parameters, because the adding a new layer usually changes many things. For that reason, as the current time I only want to make sure all the rules work correctly in single-step examples. It will be necessary to make some multi-step examples work, though it is not a good use of time to make all of them work --- there are many tasks with higher priority, as I listed in a previous email.

 

Pei

 

 

Pei

--