Fwd: [scallop-lang] Questions about how to model certain laws

[Mayur Naik]

---------- Forwarded message ---------
From: 'Joaquin' via scallop-lang <scallo...@seas.upenn.edu>
Date: Tue, May 30, 2023 at 5:11 PM
Subject: [scallop-lang] Questions about how to model certain laws
To: Scallop Lang+owners <scallop-l...@seas.upenn.edu>

Hello, it's Joaquin again. I'm trying a new approach to the problem of the CartPole. Instead of calculating the probabilities of termination given a state, I think a better approach would be to calculate the probability of being over or under the limits (defined by the rules) given the actual state.

So my program right now is:

ctx = scallopy.ScallopContext(provenance="topkproofs")

ctx.add_relation("state0", float)

ctx.add_relation("state2", float)

ctx.add_facts("state0", state0)

ctx.add_facts("state2", state2)

ctx.add_rule("terminated_ang(state2 > 0.2095 || state2 < -0.2095) = state2(state2)")

ctx.add_rule("terminated_pos(state0 > 2.4 || state0 < -2.4) = state0(state0)")

ctx.add_rule("terminated() = terminated_pos(True) or terminated_ang(True)")

ctx.run()

But when I execute the probability, it gives me the value of the state0 or state2. Is this correct?

Also, in this case, and using the NN how can I calculate the loss function?

Another approach I'm thinking about would be to calculate the probability of being over/under the limits depending on the action, but I don't know how to model it. Any ideas?

I don't know if you have read the answer to the other email, let me know what you think.

Best,

Joaquin