[Bug] Structure transformation

[maxeeem]

Describe the bug
Structure transformation is currently not handled correctly.

See this test case from OpenNARS

        'Structure transformation

        'Bright is similar to smart. 
        <bright <-> smart>. %0.90;0.90% 

        'Is bright thing a type of smart thing?
        <[smart] --> [bright]>?

        6

        'Bright thing is a type of smart thing. 
        ''outputMustContain('<[bright] <-> [smart]>. %0.90;0.90%')
        ''outputMustContain('<[smart] --> [bright]>. %0.90;0.66%')

To Reproduce
Launch console python -m pynars.Console and type

<bright <-> smart>. %0.90;0.90%
<[smart] --> [bright]>?
6

Expected behavior
The system should be able to derive the following statements:

<[bright] <-> [smart]>. %0.90;0.90%
<[smart] --> [bright]>. %0.90;0.66%

[bowen-xu]

The following rules are missing

<S <-> B>.
[S] 
|-
<[S] <-> [P]>.

<S <-> P>.
{S} 
|-
<{S} <-> {P}>.

<A <-> B>.
<A --> B>?
|-
<A--> B>.

[maxeeem]

@bowen-xu I'm not sure additional rules are necessary. This should be handled by structural inference by applying the theorems. We already have some of them https://github.com/bowen-xu/PyNARS/blob/ebff7b965b834334b3612f5431bad20057b3f4cf/pynars/NAL/Theorems/StructuralRules.py#L979

So perhaps just need to check that all of them are in place and that control mechanism correctly applies them.

[bowen-xu]

@bowen-xu I'm not sure additional rules are necessary. This should be handled by structural inference by applying the theorems. We already have some of them

https://github.com/bowen-xu/PyNARS/blob/ebff7b965b834334b3612f5431bad20057b3f4cf/pynars/NAL/Theorems/StructuralRules.py#L979

So perhaps just need to check that all of them are in place and that control mechanism correctly applies them.

Yes, the functions have alreadly been there. So we just need to link them to the rule table. See in pr https://github.com/bowen-xu/PyNARS/pull/81

https://github.com/bowen-xu/PyNARS/blob/4000e823388c2dd48cc1b4e1e2e50982b4f13f26/pynars/NARS/InferenceEngine/GeneralEngine/Rules/NAL2.py#L203-L240

[maxeeem]

@bowen-xu I guess maybe I'm a little lost and not able to see the connection between the functions. When is the theorem used? I don't see it referenced anywhere if I do a search on the whole codebase, for example equivalence_theorem4. Can you please explain how theorems and rules are connected?

And in the new rules, where does the second premise come from? [S] or {S} for example?

[bowen-xu]

@maxeeem I thought equivalence_theorem4 is a special case of implication_theorem5, so that it calls function _structural__bi_composition__0 when meeting {<S <-> P>., [S]} (as 3.0.4 does). In implication_theorem5 it calls function bi_composition, while in equivalence_theorem4 it can do the same thing (although currently it doesn't call bi_composition but is implemented by seperated code).

But maybe you're right; I should make the code more clear by directly calling equivalence_theorem4 for example.

[bowen-xu]

[S] is the name of concept [S]. It would be passed to the inference engine when there is no belief in the concept's beliefs-table.