This one should be a simple plug-and-play using a PyNARS .exe. Should I try v0.2 or the dev branch?

The demo can be simple:

• play Pong given background knowledge (requires goal derivation and handling)

or it can be complicated

• learn Pong without any background knowledge (requires goal handling, plus temporal sequence learning (eg. <A &/ ^B => C>) (ONA and OpenNARS can do this one already)

What is the current state of the implementation. Are we ready for the latter, or do we still need more coding? Can we at least pull off the former?

I prefer dev. Since some bugs have been fixed in dev.

Maybe start from the simple one? The temporal induction procedure has not been proved correct in the current implementation of PyNARS. I'm not sure if there's any other bug for the control part. If it works well in the simple case, then can focus on the temporal induction.

Maybe start from the simple one? The temporal induction procedure has not been proved correct in the current implementation of PyNARS. I'm not sure if there's any other bug for the control part. If it works well in the simple case, then can focus on the temporal induction.

Or we can create a new repository (It can be called like "PyNARS-Demo") to make progresses parallel?

Added .exe and an option to select PyNARS in ccrock4t/NARS-Pong@9113543

Will run some experiments

Here is the output log, resulting from running the current dev branch on Pong:
PyNARS_Log.txt

Currently, the system does not execute any Operations.
It does accept input goals and events. It also accepts background knowledge (implication statements).

There may be a bug. For some reason the system is outputting the Judgment as True

The Event truth values are not decaying

Input:  0.90  0.90  0.95  IN    : <{SELF}-->[good]>! :|: %1.000;0.900%
...
Input:  0.80  0.50  0.95  IN    : <{SELF}-->[good]>. :|: %1.000;0.900%
....
Input:  0.90  0.90  0.95  IN    : <{SELF}-->[good]>! :|: %1.000;0.900%
....
 1.00  0.50  0.95  OUT   : <{SELF}-->[good]>. :\: %1.000;0.900%
 0.13  0.88  0.95  OUT   : <{SELF}-->[good]>! :\: %1.000;0.900%

Any insights based on these results?

Is the long syntax of Operations (e.g., <(*, {SELF})-->deactivate>) supported? Regardless, I also tried the short version (e.g., ^deactivate) but no change in the results.

After reviewing the output log more closely, it seems like the Truth-Values of Events do not decay in the system.

Is the long syntax of Operations (e.g., <(*, {SELF})-->deactivate>) supported? Regardless, I also tried the short version (e.g., ^deactivate) but no change in the results.

At present, in PyNARS, operator need to be registered to EXE operations. I recently added a cmd called register in ConsolePlus, and you can register it through /reg deactivate eval.
(You can refer to PR #37)

After reviewing the output log more closely, it seems like the Truth-Values of Events do not decay in the system.

According to the code, it is obvious that PyNARS has not fully implemented NAL-7 at present.
I think a complete implementation of NAL-7 (such as {A. :|:, B. :|:} |- <A ==> B>.) is necessary in order to realize temporal reasoning in demo.

@ccrock4t OK. I will try to find out the bug this week.

@ARCJ137442 Yes, you're right.

Currently, pynars cannot deal with temporal induction very well. Although the inference rules have been implemented, the control mechanism on temporal induction has not been realized yet. The code on temporal induction (dealing with anticipation) in OpenNARS 3.0.4 is mainly in file ProcessAnticipation.java. Something similar should be done in pynars.