We should be automating more steps of interpretation. One potential way to do this:
Finding samples where IC-K is high/low AND that have similar hx-grads and env-grads for some particular quantity that we take the grad of (e.g. "show me samples where IC6 is high and the grads of IC8 at t-2 are similar"). I think this has a good chance of identifying groups of samples that have some obvious causal structure. And if it does, it should make statistical claims way easier.
A later issue will require adding a filter in the panel for this.
We should be automating more steps of interpretation. One potential way to do this:
Finding samples where IC-K is high/low AND that have similar hx-grads and env-grads for some particular quantity that we take the grad of (e.g. "show me samples where IC6 is high and the grads of IC8 at t-2 are similar"). I think this has a good chance of identifying groups of samples that have some obvious causal structure. And if it does, it should make statistical claims way easier.
A later issue will require adding a filter in the panel for this.