Guy Van den Broeck

Is there a workaround?

I think EVI is not supposed to be run on non deterministic circuits. It's essentially a faster version of MAR for when data and circuit have no latents

I would like general feedback on this issue, and I'm especially curious what our external users like @RenatoGeh and @eliavw think.

Thanks for the report. DaisyDiff is looking for someone to take over the codebase to address issues like this.

I was using the git version from last weekend. Unfortunately the code is quite complex and I cannot share it at the moment.

Thanks. I believe the underlying cause is that the computation graph I give to jit is simply too large (about 10,000 np operations).

I managed to isolate the code that produces the error: [bug.zip](https://github.com/google/jax/files/3190068/bug.zip) As I thought, it's probably due to the jit compiler running out of memory. The file also shows a...

The naming issue was resolved in https://github.com/Juice-jl/LogicCircuits.jl/commit/e92c417386f4b3a9089c1302148d30cbe82c6951 where condition is now conjoin in the transformations. We should still consider adding a condition operation though, and a forgetting one -- leaving...

The function after conditioning on X no longer depends on X. The function after conjoining with X still depends on it and fixes it to one value.

I'm certainly in favor, especially in the documentation strings.