Counterexample for `no_stack_with_bar_area_discrete_color` hard constraint

[light-and-salt]

I have a counterexample for the no_stack_with_bar_area_discrete_color hard constraint :P

draco/asp/hard.lp

Line 150 in d89bc0b

hard(no_stack_with_bar_area_discrete_color,E) :- mark(bar;area), channel(E,color), discrete(E), not stack(_).

[domoritz]

Adding stack doesn't hurt in this case, though, right?

I guess we can make an exception when the color field is the same as the categorical field.

[light-and-salt]

Or we can make the rule a bit more specific like

hard(no_stack_with_bar_area_discrete_color,E) 
  :- mark(bar;area), channel(E,color), discrete(E), 
  field(E,F), field(EX,FX), field(EY, FY), F != FX, F != FY,
  not stack(_).

Hmm... this is a bit verbose...

[domoritz]

I don't understand the rule. Also note that you usually want to use > instead of != To break symmetries.

[light-and-salt]

F != FX, F != FY, is my ugly way of saying that the color field is not the same as the x field and it's also not the same as the y field

I guess I could revise the rule to make it a bit more readable:

hard(no_stack_with_bar_area_discrete_color,E) 
  :- mark(bar;area), channel(E,color), discrete(E), not redundant_color(E), not stack(_).

redundant_color(E) :- channel(E,color), field(E,F), field(EX,FX), F = FX.
redundant_color(E) :- channel(E,color), field(E,F), field(EY,FY), F = FY.

[domoritz]

EX and EY are just encodings and you never say that they are the x or y encoding, though.

[light-and-salt]

Right, so it should be

hard(no_stack_with_bar_area_discrete_color,E) 
  :- mark(bar;area), channel(E,color), discrete(E), not redundant_color(E), not stack(_).

redundant_color(E) :- channel(E,color), field(E,F), channel(EX, x), field(EX,FX), F = FX.
redundant_color(E) :- channel(E,color), field(E,F), channel(EY, y), field(EY,FY), F = FY.

[domoritz]

This should be the same as this, right?

hard(no_stack_with_bar_area_discrete_color,E) 
  :- mark(bar;area), channel(E,color), discrete(E), field(E,F), field(EE,F) not channel(EE,x), not channel(EE,y), not stack(_).

[light-and-salt]

yup! looks good to me!