IBM / lemming

Human in the loop plan selection
Apache License 2.0
7 stars 1 forks source link

BUILD Views #5

Closed TathagataChakraborti closed 1 year ago

TathagataChakraborti commented 1 year ago
TathagataChakraborti commented 1 year ago

@jkeskingvillage after the first selection on the Toy Domain, the returned plan is missing an edge (the choice infos seem to be correct)

Screenshot 2023-07-03 at 20 43 06 Screenshot 2023-07-03 at 20 43 48
jkeskingvillage commented 1 year ago

Let me investigate this.

TathagataChakraborti commented 1 year ago

Also, Build Backwards seems more of an issue:

Toy Domain

It does stop at the first fork but IMO the graph needs to show the fork as well otherwise the user cannot pick.

Screenshot 2023-07-03 at 23 54 49

Travel Domain

The whole graph shows up.

Screenshot 2023-07-03 at 23 54 37
TathagataChakraborti commented 1 year ago

@jkeskingvillage I have added an extra node for build backward so that the graph does not have multiple roots

Screenshot 2023-07-04 at 22 56 24

Now to the question of what the graph should look like

Toy Domain

For Toy Domain we get this 👇 which I think (?) is correct, except that I am highlighting in the same direction as others (forward) so the highlight is wrong. On the other hand, I don't think highlighting backward help coz here the buffer edge is not actually a choice. In general, the fork leading up from the first forking node will have some non-choices, which I supposed is a disadvantage for the BUILD mode. So nothing much to do here (I will change the highlight to the forking node only).

Screenshot 2023-07-04 at 23 08 08

Travel Domain

This one does not look correct. The graph goes all the way back to the first confluence of all edges -- which I suppose is correct. However, the highlight is on the wrong node in the middle of the first fork. This is because I am taking the first among the available_for_choice marked True (as we do in the Selection View). What's the desired action here?

Screenshot 2023-07-04 at 22 56 28 Screenshot 2023-07-04 at 23 13 15
jkeskingvillage commented 1 year ago

I can sort the choice by their distance from the initial state. I will put the one with the closest to the initial condition to the first position in ChoiceInfos.