search

opennars / OpenNARS-for-Applications

General reasoning component for applications based on NARS theory.
https://cis.temple.edu/~pwang/NARS-Intro.html
MIT License
91 stars 40 forks source link

Linear Confidence Discount in Truth Fn's by Inner w2c/c2w Transform #223

Closed automenta closed 2 years ago

automenta commented 2 years ago

currently deduction.c = (T.c * B.c) * B.f how would differ from: deduction.c = w2c(c2w(T.c * B.c) * B.f)

in other words, when ever a truth function multiplies (discounts) confidence by a factor, it is possible to instead convert to evidence scale before multiplying, then convert back to confidence scale.

thus linear amounts of evidence are preserved rather than "curved" as they would if the multiplication happened directly in confidence scale

automenta commented 2 years ago

in 'weak' truth functions (like Induction), the 'weakening' step may need to be applied first, before the linear discount. i believe the opposite order, weakening applied as final step, is usually implemented, which further curves the result.

patham9 commented 2 years ago

I discussed it with Pei, will show the results of the discussion soon. In short: the treatment would be perfect if the truth evaluations would be about the same statement but aren't. when you draw the Venn Diagram you will see the issue. Will upload it when I have time.

patham9 commented 2 years ago

Closed for now but I will post the picture next week.