"ERROR! StructMemberList not implemented yet."

[fdouglis]

Since starting to play with prost a few days ago, I've encountered a few cases of RDDL features not supported in prost. One for instance was using an enum in a switch statement, which I had to convert to if (@case1) then (something) else if(@case2) .... Not a big deal.

Here, as I posted earlier with respect to a Mars Rover example, I tried running it in prost. I hit the above error: ERROR! StructMemberList not implemented yet.. This seems harder to work around. Is there any plan for adding the things marked as "not yet," particularly this one?

[fdouglis]

I'll add ERROR! Unknow special function max defined as expression. Special functions not implemented yet. to this issue.

Did I use max wrong or is it truly impossible to assign, say, x'(?foo) = max[x(?foo)-?y, 0]

[geisserf]

The max function is (as far as I know) not part of the original RDDL specification, so it is not implemented in Prost. While this is also true for the exp function, this is supported by the planner because it was required for the IPC 2014 Wildfire domain. However, supporting max should be pretty straightforward by following the implementation of the Exponentialfunction in the planner. If this would be useful for you I can see whether I have some time over the next week to implement this.

I'll have to look into the StructMemberList issue; right now I am not even sure to what RDDL construct this refers to.

[thomaskeller79]

To be honest, I also don't know what a `StructMemberList? is. Would you please give us an example?

As a general rule of thumb, if something has been described by Scott as part of (the unpublished) "RDDL 2.0", it is very likely not supported by Prost.

[fdouglis]

To be honest, I also don't know what a `StructMemberList? is. Would you please give us an example?

As a general rule of thumb, if something has been described by Scott as part of (the unpublished) "RDDL 2.0", it is very likely not supported by Prost.

Thanks for expanding on the RDDL 2.0 disconnect, that helps.

In multiagent_mars_rover.rddl, in the rddlsim examples, there is the following line:

PICT_POS(p5) = <@x:15.0, @y:-15.0>;

I think that is an example of what triggered that error message.

I wasn't aware that RDDL2.0 is "unpublished" -- from the dates on the things I saw about it, I thought it had been around for some time. But then, as I've said, I'm new to all this. I can't actually find any documentation on RDDL2 specifically, so I'll have to buy the unpublished part, but the rddl2 files in the rddlsim git repo are 7 years old. Color me confused.

[fdouglis]

The max function is (as far as I know) not part of the original RDDL specification, so it is not implemented in Prost. While this is also true for the exp function, this is supported by the planner because it was required for the IPC 2014 Wildfire domain. However, supporting max should be pretty straightforward by following the implementation of the Exponentialfunction in the planner. If this would be useful for you I can see whether I have some time over the next week to implement this.

max would indeed be useful (and I assume you'd provide min at the same time). Thanks.

[geisserf]

@fdouglis Could you provide me a minimal example file that uses min/max?

[fdouglis]

I don't know about an example file, since I didn't actually successfully implement it, but I can give you an example. I wanted to say that a connects to b, connects to c, and so on, and to say that something going from a to x should go over the minimal path (the sum of the hops between them). I wound up having to precompute the total hops and the best direction from any point to any other point, but for anything of any interesting size, that blows up. As another example, imagine you can send something from a to x or y, and from b to x or y. If a is closer to x and b closer to y, you want a higher reward for that assignment.

IIRC, the specific max example I was trying to do at the time was using continuous arithmetic, which I then realized also wasn't possible.