search

erindb / causal_simulations

2 stars 1 forks source link

Continuous parameters #6

Closed erindb closed 5 years ago

erindb commented 5 years ago

Implement continuous velocities in the 1D, 2-ball, 2-position case. Does this reduce the endorsement of the video where B follows A?

5d148105-eea0-d6a7-84f1-9c5a6996e793

The issue here was that if A had moved in the opposite direction or if it had stayed still, then B's movement would definitely have changed. But in the continuous case, there are lots of params where that wouldn't be the case. If A slows down only a little bit, B will continue the way it has been.

Note that interpreting the counterfactual as "if A hadn't existed" would also potentially help with this, but it's maybe equivalent to A going in a wildly different direction.

erindb commented 5 years ago

Does it make sense to use variational for this?

erindb commented 5 years ago

nope, just use gaussian noise and an observe statement

erindb commented 5 years ago

state space is to large...... probably better to discretize at finer and finer resolution.....