PSL assumption

[Nouf Bindris]

Hi

please,

1-

Closed predicates follow the closed world assumption

and 

Open predicates follow the open-world assumption?

2-

in the PSL program, we only define the Closed predicates set?

Thank you so much for help.

[Eriq Augustine]

Hey Nouf,

1 - Closed predicates follow the closed world assumption and  Open predicates follow the open-world assumption?

Yes.

But remember that under non-lazy inference, atoms from open predicates must be fully specified in data files.

2 - in the PSL program, we only define the Closed predicates set?

PSL must know about all predicates.

So if you tell PSL about the closed predicates, then it also knows about the open predicates.

Of course, this is only done in the Java/Groovy interface.

In the CLI/Python all predicates are defined as open/closed.

-eriq

--
You received this message because you are subscribed to the Google Groups "PSL Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to psl-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/psl-users/59635876-8837-4ec2-af23-e188d6bf6f80%40googlegroups.com.

[Nouf Bindris]

thank you

in the closed predict the atoms that are unknown get the value 0

and the known predict it could take any value [0,1]

but what is the default value for the open predicted atoms if they are unknown and if thy known ]?

[Eriq Augustine]

Non-observed atoms from open predicates can take several different initial values depending on the configuration value: "admmreasoner.initialconsensusvalue".

https://github.com/linqs/psl/blob/develop/psl-core/src/main/java/org/linqs/psl/reasoner/admm/ADMMReasoner.java#L95

By default, it takes a random value.

Other possible values are zero and whatever value the atom is initially set at.

However, for the most part this value should not matter (unless you are deep in the optimization weeds).

Since the PSL objective is convex, choice of starting value will really only change how many iterations it takes to get to the same objective value.

-eriq

--

You received this message because you are subscribed to the Google Groups "PSL Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to psl-users+...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/psl-users/1d2411f3-b61e-46e7-bd00-70714caa2e3f%40googlegroups.com.